&EPA
United Sates
Enviromiwilal PratecSwi
Agency
Revised Technical Support Document:
National-Scale Assessment of Mercury Risk
to Populations with High Consumption of
Self-caught Freshwater Fish
In Support of the Appropriate and Necessary
Finding for Coal- and Oil-Fired Electric
Generating Units
-------
EPA-452/R-11-009
December 2011
Revised Technical Support Document: National-Scale Assessment of
Mercury Risk to Populations with High Consumption of Self-caught
Freshwater Fish
In Support of the Appropriate and Necessary Finding for Coal- and Oil-
Fired Electric Generating Units
U.S. Environmental Protection Agency
Office of Air Quality Planning and Standards
Health and Environmental Impacts Division
Research Triangle Park, North Carolina
11
-------
DISCLAIMER
This document has been prepared by staff from the Office of Air Quality Planning and
Standards, U.S. Environmental Protection Agency. Any opinions, findings, conclusions, or
recommendations are those of the authors and do not necessarily reflect the views of the EPA.
Questions related to this document should be addressed to Dr. Zachary Pekar, U.S.
Environmental Protection Agency, Office of Air Quality Planning and Standards, C504-06,
Research Triangle Park, North Carolina 27711 (email: pekar.zachary@epa.gov).
in
-------
Table of Contents:
Executive Summary viii
1 Review of Analysis Approach 1
1.1 Summary of Significant SAB Peer Review Panel Recommendations and Revisions
Reflected in this Revised Mercury Risk TSD 2
1.2 Purpose and Scope of Analysis 5
1.3 Overview of Risk Metrics and the Risk Characterization Framework 10
1.4 Overview of Analytical Approach 13
1.4.1 Specifying the spatial scale of watersheds 16
1.4.2 Characterizing measured fish tissue Hg concentrations at the watershed level 17
1.4.2.1 Projecting 75th percentile fish tissue Hg concentrations for the 2016 scenario . 31
1.4.3 Defining subsistence fisher scenarios to include in the analysis 31
Identify subsistence fisher populations 32
Assess where the subsistence fisher populations might be active 33
1.4.4 Estimating total fish consumption-related MeHg exposure (2016 scenario) 41
1.4.5 Estimating risk (RfD-based hazard quotient) (2016 scenario) 43
1.4.6 Estimation of U.S. EGU-attributable risk (2016 scenario) 43
1.4.6.1 Mercury Maps analysis 44
1.4.6.2 Additional research supporting the proportionality assumption and examining the
issue of temporal response 46
1.4.6.3 CMAQ mercury deposition modeling 47
1.5 Differences between the 2005 Section 112(n) Revision Rule analysis and the current
analysis in support of the Appropriate and Necessary Finding 48
1.6 Detailed Example Calculation (watershed-level riskHQ) 50
2 Discussion of Analytical Results 53
2.1 Key design elements to consider when reviewing the risk assessment results 53
2.2 Mercury Deposition from U.S. EGUs as Modeled Using CMAQ 54
2.3 Fish Tissue Mercury Concentrations 65
2.4 Comparing Patterns of Hg Deposition with Hg Fish Tissue Data for the 3,141
Watersheds Included in the Risk Assessment 74
2.5 Overview of Risk Estimates 80
2.5.1 Overview of percentile risk estimates (2016 scenario) 80
2.5.2 Overview of number and percentage of watersheds with populations potentially at-
risk due to U.S. EGU mercury emissions (2016 scenario) 83
2.6 Sensitivity Analyses 87
2.7 Discussion of key sources of variability and uncertainty 91
3 Summary of Key Observations 110
Appendices: Additional Technical Detail on Modeling Elements and Presentation of
Supplemental Risk Estimates 117
Appendix A. Technical Approach Used in Modeling IQ Loss 117
Appendix B. Supplemental Risk Estimates (IQ loss estimates) 118
Appendix C. SAB Mercury Panel Peer Review Letter: Review of EPA's Draft National-Scale
Mercury Risk Assessment 120
IV
-------
List of Tables
Table 1-1 Summary Statistics for the 2011 MFT HUC-level Data 25
Table 1-2 Summary of 75th Percentile Hg Concentrations in Fish Tissue Samples by Number of
Observations and Number of Sites per HUC 27
Table 1-3 Summary of 50th Percentile Hg Concentrations in Fish Tissue Samples by Number of
Observations and Number of Sites per HUC 29
Table 1-4 Comparison of HUC-Level Fish Tissue Hg Statistics for (a) Fish Tissue Dataset with
Fish >7 Inches and (b) Dataset with All fish Lengths Included 31
Table 1-5 Spatial Extent and Number of HUCs Reflected in Risk Modeling for the Female
subsistence fish consumer Scenarios Included in the Risk Assessment 32
Table 1-6 Fish consumption rates and additional behavior-related information for subsistence
populations included in the analysis 38
Table 2-1 Comparison of total and U.S. EGU-attributable mercury deposition (|ig/m2) for the
2005 and 2016 scenarios.* 64
Table 2-2 Comparison of percent of total mercury deposition attributable to U.S. EGUs for 2005
and 2016.* 64
Table 2-3 Comparison of percent reduction of total mercury deposition, and U.S. EGU-
attributable deposition, based on comparing the 2016 scenario against the 2005 scenario.* 64
Table 2-4 Comparison of HUC-Level Fish Tissue Mercury Concentrations Across Datasets
Used in the March Version and Current Version of the Risk Assessment 73
Table 2-5 Comparison of total and U.S. EGU-attributable Hg fish tissue concentrations
(including % change) for the original fish tissue dataset and 2016 scenarios* 73
Table 2-6 Percentile risk estimates for the full set of female subsistence fish consumer scenarios
included in the analysis (2016 scenario) (for both total and U.S. EGU incremental RfD-based
HQ)* 81
Table 2-8 Watersheds with potentially at-risk populations based on consideration for risk based
on U.S. EGU mercury deposition and resulting exposure considered alone, without taking into
account other sources of mercury deposition (2016 scenario) 85
Table 2-9 Combination of watersheds with potentially at-risk populations based on either
consideration for (a) U.S. EGU percent contribution to total risk OR (b) risk when U.S. EGU
mercury deposition is considered alone, without taking into account deposition from other
sources (2016 scenario) 85
Table 2-10 Reflecting the March version of the risk assessment - combination of watersheds
with potentially at-risk populations based on either consideration for (a) U.S. EGU percent
contribution to total risk OR (b) risk when U.S. EGU mercury deposition is considered alone,
without taking into account deposition from other sources (2016 scenario) 86
Table 2-11 Sensitivity analysis results presented as: watersheds with potentially at-risk
populations based on U.S. EGUs making a specified contribution to total risk (2016 scenario) 88
Table 2-12 Sensitivity analysis results presented as: watersheds with potentially at-risk
populations based on consideration for U.S. EGU-attributable HQ risk (2016 scenario) (risk
considering U.S. EGU Hg deposition before considering other sources of Hg deposition) 89
Table 2-13 Sensitivity analysis results presented as: Combination of watersheds with potentially
at-risk populations based on either consideration for (a) U.S. EGU percent contribution to total
risk OR (b) risk when U.S. EGU mercury deposition is considered alone, without taking into
account deposition from other sources (2016 scenario) 90
-------
Table 2-14 Key sources of variability associated with the analysis and degree to which they are
reflected in the design of the analysis 93
Table 2-15 Key sources of uncertainty associated with the analysis, the nature of their potential
impact on risk estimates, and degree to which they are characterized as part of the analysis 96
List of Figures
Figure 1-1. 2-StageRisk Characterization Framework 13
Figure 1-2 Flow Diagram of Risk Analysis (for the 2016 air quality scenario) Including Major
Analytical Steps and Associated Modeling Elements 14
Figure 1-3 Diagram Illustrating Step-wise Procedure Used to Develop 2010 Mercury Fish
Tissue (MFT) DatasetUsed in the 2010 National-Scale Mercury Risk Assessment 21
Figure 1-4 Diagram Illustrating Step-wise Procedure Used to Develop the Augmentation
Mercury Fish Tissue (MFT) Dataset 22
Figure 1-5 Diagram Illustrating Step-wise Procedure Used to Combine the 2010 MFT and
Augmentation MFT Datasets 23
Figure 1-6 Diagram Illustrating Number of HUC12s with Fish Tissue Mercury Data (for 2010
MFT, Augmentation MFT and the Combined 2011 MFT Datasets) 26
Figure 1-7 Histogram Characterizing Frequency of Sample Sizes Across HUCs Included in the
Risk Assessment (illustrates fraction of HUCs with small sample size of 1-2) 28
Figure 1-8 LOESS (locally-weighted scatter plot smoothing) Least-Square Regression of 75th
Percentile HUC-level Fish tissue Hg Levels Against HUC-level Sample Size 29
Figure 1-9 Sample Calculation for watershed-level Risk HQ 52
Figure 2-1 Total Mercury Deposition by HUC (|ig/m2) for the 2005 Scenario 56
Figure 2-2 Total Mercury Deposition by HUC (|ig/m2) for the 2016 Scenario 57
Figure 2-3 U. S EGU-Attributable Mercury Deposition by HUC (|ig/m2) for the 2005 Scenario
58
Figure 2-4 U.S EGU-Attributable Mercury Deposition by HUC (|ig/m2) for the 2016 Scenario59
Figure 2-5 Mercury Wet Deposition by 12km Grid Cell (|ig/m2) for the 2005 Scenario 60
Figure 2-6 Mercury Dry Deposit!on by 12km Grid Cell (|ig/m2) for the 2005 Scenario 61
Figure 2-7 Mercury Wet Deposition by 12km Grid Cell (|ig/m2) for the 2016 Scenario 62
Figure 2-8 Mercury Dry Deposit!on by 12km Grid Cell (|ig/m2) for the 2016 Scenario 63
Figure 2-9 Set of 3,141 HUC12s with Fish Tissue Mercury Data Used in the Risk Assessment*
67
Figure 2-10 Fish Tissue Measurement Sampling Frequency for HUCs with Fish Tissue Data
Included in the Risk Assessment 68
Figure 2-11 Total Fish Tissue Mercury Concentrations for 2005 Scenario (HUC12-level 75th
percentile values, ppm) 69
Figure 2-12 Total Fish Tissue Mercury Concentrations Projected for 2016 Scenario (HUC12-
level 75th percentile values, ppm)* 70
Figure 2-13 EGU-Attributable Fish Tissue Mercury Concentrations for 2005 Scenario 71
Figure 2-14 EGU-Attributable Fish Tissue Mercury Concentrations Projected for 2016 Scenario
72
VI
-------
Figure 2-15 Comparison of Locations of Watersheds with Fish Tissue Hg Data* with Pattern of
U.S. EGU-attributable Hg Deposit!on (2005 scenario) 76
Figure 2-16 Comparison of Locations of Watersheds with Fish Tissue Hg Data* with Pattern of
U.S. EGU-attributable Hg Deposit!on (2016 scenario) 77
Figure 2-17 For the 2005 Scenario, Scatter Plot of Hg Fish Tissue Concentrations Versus Total
Hg Deposition for the 3,141 Watersheds Included in the Risk Assessment 78
Figure 2-18 Cumulative distribution plots of U.S. EGU-attributable Hg deposition over the
3,141 watersheds used in modeling the high-end female consumer population as contrasted with
all 88,000 watersheds (2016 Scenario) 78
vn
-------
Executive Summary
The Technical Support Document is a revised version of the TSD for EPA's National-
scale Mercury Risk Assessment completed in March 2011. EPA used the previous version of
this report as support for the 2011 finding that it is appropriate and necessary to regulate coal and
oil-fired electric utility steam generating units in the United States (U.S. EGUs), pursuant to
Section 112(n)(l)(A) of the Clean Air Act.
EPA commissioned a formal peer review of this assessment during the public comment
period for the regulation through the EPA Science Advisory Board (SAB), which provides
independent advice and peer review to EPA's Administrator on the scientific and technical
aspects of environmental issues. The SAB established a 22-member panel with representation
from academic institutions, industry, federal agencies, and state governments. The SAB
supported the overall design of the risk assessment, confirmed EPA's analytical assumptions,
and concluded that the risk assessment should provide "an objective, reasonable, and credible
determination" of the potential public health hazard. The SAB made many recommendations for
improving this TSD, which SAB organized into three general themes: (1) improve clarify of the
document regarding methods and presentation of results, (2) expand discussion of sources of
variability and uncertainty, and (3) de-emphasize IQ loss as an endpoint. EPA has responded to
the peer review by substantially revising this TSD.
Purpose and Scope of Analysis
The goal of this assessment was to determine whether mercury emitted from U.S. EGUs
poses a potential public health hazard. Therefore, we have designed this risk assessment as a
screening analysis focused on identifying watersheds where there is a public health hazard
attributable to U.S. EGU mercury deposition. Mercury emitted from U.S. EGUs, depending on
the form of mercury emitted and other factors, can deposit locally and regionally in U.S.
waterbodies, as well as contribute to the global pool of mercury, where it can be transported and
eventually deposited around the world. This deposited mercury is transformed into
methylmercury (MeHg) by microorganisms and then bioaccumulates as MeHg in fish. The
primary pathway of concern from a public health standpoint is consumption of mercury-
contaminated fish by women of child bearing age (since mercury can stay in the system for some
time, both women who are pregnant or about to be pregnant are of concern.) Depending on the
level of prenatal exposure, children born to those women may then experience a range of
neurodevelopmental effects including decrements on a number of neuropsychological measures.
We focused on consumption from inland freshwater waterbodies in the U.S rather than
estuarine or marine waterbodies because the U.S. EGU-attributable mercury deposition is a
larger fraction of total mercury deposition, particularly for waterbodies with elevated U.S. EGU-
attributable mercury deposition such as the Ohio River Valley. We focused on subsistence
fishing populations because they typically have substantially higher consumption rates of self-
caught fish than recreational fishers and therefore, experience higher risk. Because we do not
have data available on the distribution of subsistence fishing populations in all watersheds in the
U.S., we modeled a hypothetical female subsistence consumer at those watersheds where we
have fish tissue data and where we believe subsistence fishing activity has the potential to occur.
Vlll
-------
Risk Assessment Methods
We followed a 10-step process to estimate risk to female subsistence consumers.
1. Model total and U.S. EGU-attributable mercury deposition for the continental U.S.
2. Interpolate mercury deposition to watersheds
3. Estimate 75th percentile fish tissue mercury concentrations at watersheds with fish tissue
data
4. Project 75th percentile fish tissue mercury concentrations for the 2016 scenario, which
reflects projected Hg deposition without regulation of mercury emissions from U.S.
EGUs
5. Define female subsistence consumer scenarios
6. Identify watersheds with subsistence fisher population activity
7. Define self-caught fish consumption rates for the subsistence scenarios
8. Estimate total fish consumption-related MeHg exposure
9. Estimate total MeHg risk
10. Estimate U.S. EGU-attributable risk
We assume a proportional relationship between changes in mercury deposition and
changes in fish tissue mercury concentrations. This assumption, supported by the Mercury Maps
assessment and other scientific literature, is used to estimate the fraction of U.S. EGU-
attributable MeHg in fish and to project fish tissue data to 2016. We use the 75th percentile fish
tissue sample because we believe it is likely that subsistence fishers may target larger fish (with
somewhat higher MeHg levels) to supplement family meals. Although we can only calculate risk
for the 4% (3,100) of the 88,000 watersheds in the continental U.S. for which we have fish tissue
mercury data, these data cover 48 states including many states with high levels of mercury
deposition and fish tissue mercury concentrations. We excluded watersheds with potentially
significant non-air sources of mercury.
In estimating risk for the most comprehensive scenario reflecting exposures for women of
childbearing age with subsistence level consumption of freshwater self-caught fish, we apply the
scenario to all watersheds with fish tissue mercury data, reflecting the potential for these
populations to obtain and consume fish from any watershed. To estimate this risk, we use the
EPA's MeHg reference dose (MeHg RfD), which identifies a level of exposure above which
there is the risk of adverse health effects. We compare modeled MeHg exposure levels against
the RfD to generate a hazard quotient (HQ). An HQ above one represents a potential exposure
above the MeHg RfD and EPA, supported by the SAB, considers such an exposure to represent a
public health hazard. We use two risk metrics at the watershed level: (a) number and percent of
watersheds modeled where the total mercury-based HQ>1 and where U.S. EGU-attributable
mercury deposition contributes at least 5% of the risk and (b) number and percent of watersheds
modeled where U.S. EGU-sourced mercury considered alone, without taking account deposition
from other sources, results in an HQ>1. The combination of these two estimates provides the
ix
-------
total number and percent of watersheds where female subsistence consumers are potentially at-
risk.
We have also modeled an additional set of female subsistence consumer scenarios
focused on different socioeconomic (SES) groups (e.g., low income Blacks in the Southeast, low
income Hispanics, Vietnamese). Each of these female subsistence consumer scenarios is
supported by scientific data on SES specific high-end subsistence-level fish consumption rates
for self-caught fish from inland freshwater bodies. We model watersheds for each of these
scenarios if there are at least 25 individuals from the SES group in close proximity to the
watershed.
Response to SAB Recommendations
In response to the SAB review, we have substantially modified this TSD. Key
modifications include: (a) clarifying that the analysis is watershed-focused and designed to
assess risk for female subsistence consumers (and not a representative characterization of risk for
recreational anglers), (b) refining and improving the technical presentation in general, including
addition of a sample calculation figure clarifying how the watershed-level risk estimates are
generated, (c) expanding the discussion of how the fish tissue mercury database was developed,
(d) incorporating additional fish tissue mercury data for states with high levels of U.S. EGU
mercury deposition, (e) verifying the linkages between type offish tissue mercury measured,
type offish consumption rates used and application of a cooking/preparation adjustment factor,
(f) inclusion of sensitivity analysis involving median watershed-level fish tissue mercury levels,
(g) expanded discussion of key sources of variability and uncertainty, and (h) de-emphasized the
IQ loss estimates and moved those estimates to an appendix In response to public comments, we
have also moved the previously titled "Hotspot Analysis" to a separate TSD titled "Potential for
Excess Local Deposition of Mercury in Areas near U.S. EGUs".
Key Observations
• Based on a combination of the two risk metrics, and reflecting consumption rates ranging
from the 90th percentile to the 99th percentile, we estimate that 22% to 29% of the
watersheds modeled have populations that are potentially at-risk due to U.S. EGU Hg
emissions in 2016.
o Based on U.S. EGU-attributable deposition considered alone, without taking into
account other sources of deposition, 10% of the modeled watersheds have
populations that are potentially at-risk in 2016 (based on the 99th percentile
consumption rate) (2016 is the year when regulations under Section 112 of the
Clean Air Act would need to be implemented for EGUs).
o In 2016, 24% of the watersheds have populations that are potentially at-risk due to
total deposition with at least 5% of that total deposition attributable to U.S. EGUs
(based on the 99th% percentile consumption rate).
• These risk results reflect U.S. EGU-attributable deposition. In 2016, we estimate that for
watersheds modeled in the risk assessment, U.S EGUs contribute up to 16% of total
mercury deposition and related fish tissue mercury concentrations. On average, for these
-------
modeled watersheds, U.S. EGUs contribute 3% of total mercury deposition and related
fish tissue mercury concentrations.
• Reducing U.S. EGU-attributable mercury will reduce the magnitude of the risk from total
mercury exposure. However, a large fraction of modeled watersheds would still have
populations at-risk due to deposition from global sources of mercury, although the degree
of risk would be diminished.
XI
-------
1 Review of Analysis Approach
1.1 Introduction
The EPA completed a national-scale risk assessment for mercury (Hg) that informed the
March 2011 finding that it is appropriate and necessary to regulate electric utility steam
generating units in the United States (U.S. EGUs), pursuant to Section 112(n)(l)(A) of the Clean
Air Act (CAA). That-risk assessment is documented in the Technical Support Document:
National-Scale Mercury Risk Assessment Supporting the Appropriate and Necessary Finding for
Coal and Oil-Fired Electric Generating Units (U.S.EPA, 201 la), hereafter referred to as the
"Mercury Risk TSD (March publication)" or "March TSD". In making that finding, EPA
determined that this assessment should be peer-reviewed, and that the results of the peer review
and any EPA response to them would be published before the final rule. EPA conducted a
formal peer review through the EPA Science Advisory Board, which provides independent
advice and peer review to EPA's Administrator on the scientific and technical aspects of
environmental issues. The SAB established a 22-member peer review panel with representation
from academic institutions, industry, federal agencies, and state governments. The peer review
charge questions and results of the peer review provided in letter form are publically available on
the EPA Science Advisory Board website,1 and are included in this document (see Appendix C).
EPA reviewed the SAB comments and recommendations for strengthening the national-
scale risk assessment and developed a revised version of the risk assessment reflecting
implementation of a number of the recommendations made by the SAB. This document
(hereafter referred to as the "Mercury Risk TSD (revised)" or "Revised TSD" describes the
technical approach used in completing the revised version of the risk assessment. While the
underlying technical approach used in assessing risk remains essentially the same, some
elements of the risk model, including some data inputs, have been modified and the presentation
and discussion of methods and risk estimates has been significantly revised in accordance with
SAB recommendations.
The remainder of this revised TSD is organized as follows. We begin this section with a
summary of the significant recommendations provided by the SAB peer review panel, and an
overview of the changes to the technical approach made in response to specific SAB
recommendations (section 1.1). We then provide a description of the purpose and scope of the
analysis (section 1.2). Next, we provide an overview of the risk metrics generated and the 2-
stage risk characterization framework used to help interpret the risk estimates in a policy-
relevant context (section 1.3). In section 1.4, we provide an overview of the analytical approach
used in the analysis, with subsections addressing specific elements of the analysis. Section 1.5
describes differences between the current risk assessment and the assessment completed in 2005
(the 2005 Section 112(n) Revision Rule analysis). Section 1.6 presents a sample calculation that
1 Link to the SAB letter (which includes the charge questions as an appendix):
http://vosemite.epa.gov/sab/sabproduct.nsf/02ad90bl36fc21ef85256eba00436459/BCA23C5B7917F5BF8525791A
0072CCAl/$File/EPA-SAB-ll-017-unsigned.pdf
1
-------
walks the reader through the process used to generate risk estimates in the analysis. This detailed
example calculation should help the reader to see how the various inputs and intermediate
modeling calculations come together in generating watershed-level hazard quotient (HQ)-based
risk estimates. Section 2 describes analytical results of the risk assessment including both
intermediate modeling results and the risk estimates that are generated (a detailed overview of
the subsections in section 2 is presented at the beginning of that section). Section 3 presents a
summary of key observations from the analysis.
1.1 Summary of Significant SAB Peer Review Panel Recommendations and
Revisions Reflected in this Revised Mercury Risk TSD
In response to the comments and recommendations by the SAB as part of their review of
the March TSD, EPA revised the technical approach and documentation for the national-scale
Hg risk assessment. These enhancements to the technical approach are briefly described below,
along with a summary of the original SAB comment in italics.2
• The watershed-focus of the risk assessment should be clearly stated. We revised the
technical approach section to more clearly state that the analysis is focused on assessing
risk for subsistence-level consumers of self-caught fish evaluated at the watershed-level.3
Specifically, we are modeling risk for a set of female subsistence fish consumer scenarios
at those waterbodies where (a) we have measured fish tissue mercury (Hg) data and (b) it
is reasonable to assume that high-end fishing activity could occur. We emphasize the
point that the analysis is not a representative population-weighted assessment of risk.
Rather, it is based on evaluating exposure scenarios.
• Because IQ does not fully capture the range of neurodevelopmental effects associated
with Hg exposure, this endpoint should be deemphasized (and covered in an appendix)
and primary focus should be placed on the MeHg RfD-based hazard quotient metric. We
modified the structure of the Revised TSD accordingly, including discussion of
alternative neurodevelopemental endpoints (besides IQ) reflected in the MeHg RfD (see
section 1.3).
• Clarify the rationale for using an HQ at or above 1.5 as the basis for selecting potentially
impacted water sheds. We revised this discussion accordingly (see section 1.4.5).
• Additional detail should be provided on the characteristics of the fish tissue Hg dataset,
including its derivation and the distribution of specific attributes across the dataset (e.g.,
number offish tissue samples and number of different waterbodies reflected in the
2 We focused this section on those SAB comments that resulted in more significant revisions to the TSD or the risk
assessment itself. Those comments that are more editorial in nature, or address less substantial technical elements of
the analysis, are not covered here, although we addressed those comments within the Revised TSD. The charge
questions for the peer review together with the SAB letter response provided as part of the peer review are included
in Appendix C (the SAB letter includes the original EPA charge questions as an attachment).
3 The assessment focuses on women of child-bearing age who consume subsistence-levels of fish that they catch or
are caught by relatives/acquaintances and shared with them (hereafter referred to as female subsistence fish
consumers). As discussed in section 1.4 and 1.4.3, we model a number of female subsistence fish consumer
scenarios that are differentiated to provide coverage for different socio-economic status (SES) differentiated groups.
In this context, "subsistence" refers to individuals who rely on noncommercial fish as a major source of protein in
their diet (U.S. EPA, 2000). For purposes of this risk assessment, we have interpreted this as representing self-
caught fish consumption ranging from a fish meal (8 ounce) every few days to a large fish meal (12 ounces or more)
every day.
-------
percentiles provided for a given hydrologic unit code (HUC)4, number of species
reflected across HUCs). We included additional figures and tables describing the
derivation of the HUC-level fish tissue Hg dataset including the filtering steps to derive
the fish tissue dataset and to generate the HUC-level percentile estimates (see section
1.4.2). In addition, we included tables summarizing the distribution of key attributes
within that dataset highlighted by the SAB (e.g., distribution offish tissue sample size
and number of species across the HUC-level estimates). We also provided a table that
contrasts fish tissue Hg concentrations for fish greater than 7 inches in length with
concentrations based on all fish, without filtering on size (i.e., the dataset used in the risk
assessment, versus all fish combined).
Determine whether there is additional (more recent) fish tissue data for key states
including Pennsylvania, New Jersey, Kentucky and Illinois where U.S. EGUs Hg
deposition may be more significant. We expanded the fish tissue dataset by incorporating
additional (newer) fish tissue data from the National Listing of Fish Advisory (NLFA)
(which included additional data for four states including MI, NJ, PA, and MN). We also
obtained additional data for Wisconsin (see section 1.4.2).
Regarding the fish tissue dataset used in the risk assessment, clarify which species ofHg
is reflected in the underlying samples and discuss the implications of differences across
states in sampling protocols in introducing bias into the analysis. We clarified that in
most cases, the fish tissue is measured for total Hg. Furthermore, based on the literature,
it is reasonable to assume that more than 90% offish tissue Hg is methylmercury
(MeHg). Therefore, we incorporated a Hg conversion factor of 0.95 into our exposure
calculations to account for the fraction of total Hg that is MeHg in fish (see section
1.4.4). We also expanded the discussion of uncertainty (see Table 2-15 entry B in section
2.7) to include the potential for different sampling protocols across states to introduce
bias into the risk assessment.
Expand the uncertainty discussion related to the screening out of HUCs from the risk
assessment to address (a) uncertainty in characterizing significant non-air sources ofHg
loading reflected in the Toxics Release Inventory (TRl) and (b) failure to consider release
of Hg from larger urban areas (e.g., sanitary sewer discharges) in screening out HUCs.
We expanded the uncertainty discussion presented in Table 2-15 entry D of section 2.7 to
address these two sources of uncertainty related to the exclusion criteria for HUCs.
While there is support for the use of the 75th percentile fish tissue Hg value in the risk
assessment, there is concern that the low sampling rates reflected across the HUCs may
low-bias the 75th percentile estimates. As noted above, we provided additional
description of the fish tissue dataset including distribution of sample sizes and fish
species across the HUCs, which includes an improved discussion of uncertainty and
potential bias related to the derivation and use of the 75th percentile fish tissue levels (see
Table 2-15 entry C in section 2.7, as well as the discussion in section 1.4.2).
Include a sensitivity analysis based on use of the median fish tissue Hg value (as
contrasted with the 75th percentile value) in generating risk estimates: We also included a
sensitivity analysis that used the 50th percentile watershed-level fish tissue Hg
4 For the assessment we have selected Hydrologic Unit Code 12 (HUC12) watersheds as the spatial unit of analysis
(see section 1.4.1). Throughout this revised TSD, unless otherwise stated, the terms "watershed" and "HUC" are
used interchangeably to refer to HUC12's.
-------
concentration, in addition to the 75th percentile value used in the core analysis (see
section 2.6)
The SAB was generally supportive of the consumption rates used, but recommended that
EPA expand its discussion of caveats associated with the fish consumption rates used in
the analysis (e.g., high-end consumption rates for South Carolina reflect small sample
sizes, the consumption surveys underlying the studies are older and behavior may have
changed, consumption rates may reflect seasonally even if they are expressed in annual
terms). We expanded the discussion of uncertainty related to the fish consumption rates
to address the caveats identified by the SAB (see Table 2-15 entries F and G in section
2.7).
The EPA needs to verify whether the consumption rates are daily values expressed as
annual averages and whether they are "as caught" or "as prepared" (these factors have
important implications for the exposure calculations). As noted in section 1.4.4, we
carefully reviewed the studies underlying the fish consumption rates used in the risk
assessment and verified that the rates are annual-average daily consumption rates and that
they represent as prepared estimates. We also expanded the explanation of the exposure
calculations to more completely describe the exposure factors and equation used to
generate the average daily MeHg intake estimates for the female subsistence fish
consumer scenarios (see section 1.4.4).
Need to explain exclusion offish smaller than 7 inches in length from analysis. We
described the rationale for the 7-inch cutoff for edible fish used in the risk analysis (see
section 1.4.2).
The SAB was generally supportive of the approach used for identifying HUCs with the
potential for subsistence activity; however, they did recommend that we identify the
number of HUCs excluded from the analysis due to this criterion. We added a table to
illustrate the number of HUCs with fish tissue Hg data used to model risk for each of the
female subsistence fish consumer scenarios (see section 1.4.3, Table 1-5).
Enhance the discussion of the proportionality assumption linking Hg deposition and fish
tissue Hg concentrations, including more recent studies supporting the proportional
relationship between changes in Hg deposition and changes in MeHg fish tissue levels.
We expanded our discussion of uncertainly on the proportionality assumption and added
citations for the more recent studies supporting the proportionality between changes in
Hg deposition and changes in fish tissue Hg concentrations (see Table 2-15 entries I
through M in section 2.7 for additional uncertainty discussion and section 1.4.6 for
discussion of newer literature supporting the proportionality assumption).
Given that there are published studies comparing measurements of wet deposition with
CMAQ-based estimates, the presentation of Hg deposition estimates should include maps
of wet and dry Hg deposition in addition to the total Hg deposition results used in the risk
assessment.: We expanded the presentation of Hg modeling results to include maps of
wet and dry deposition for both the 2005 and 2016 scenarios (see section 2.2).
The discussion of the concentration-response (C-R) function used in modeling IQ loss,
should be expanded to include coverage for the potential masking effect of MeHg-related
IQ loss due to nutrients (polyunsaturatedfatty acids - PUFAs) in fish. This is particularly
relevant iflQ loss functions derived from saltwater fish consumption are used to model
risk associated with freshwater fish consumption. We expanded the discussion of
-------
uncertainty on modeling IQ loss to address this issue (see Table B-2 entry B in Appendix
B).
Additional sources of variability should be discussed in terms of the degree to which they
are reflected in the design of the risk assessment and the impact that they might have on
risk estimates. These include:
o The geographic patterns of populations of female subsistence fish consumers,
including how this factor interacts with the limited coverage we have for
watersheds with our fish tissue Hg data.
o The protocols used by states in collecting fish tissue Hg data.
o Fisher body weights and the impact that this might have on exposure estimates.
o Preparation and cooking methods which effects the conversion offish tissue Hg
concentrations (as measured) into "as consumed" values.
We expanded the discussion of sources of variability presented in Table 2-14 (in section
2.7) to more fully address these sources of variability as requested.
Additional sources of uncertainty should be discussed in terms of their potential impact
on risk estimates. These include:
o Emissions inventory (including the non-EGU segment) used in projecting total
and U.S. EGU-attributable Hg deposition. This includes the projection of
reduction in U.S. EGU emission for the 2016 scenario.
o Air quality modeling with CMAQ including the prediction of future air quality
scenarios.
o A bility of the Mercury Maps-based approach for relating Hg deposition to MeHg
in fish to capture Hg hotspots.
o The limited coverage that we have with fish tissue Hg data for watersheds in the
U.S. and implications for the risk assessment.
o The preparation factor used to estimate "as consumed" fish tissue Hg
concentrations.
o Proportionality assumption used to relate changes in Hg deposition to changes in
fish tissue Hg concentrations at the watershed-level.
o Characterizing the spatial location of female subsistence fish consumer
populations (including degree to which these provide coverage for high-
consuming recreational fishers).
o Application of the RfD to low SES populations and concerns that this could low-
bias the risk estimates.
We expanded the discussion of sources of uncertainty presented in Table 2-15 (in section
2.7) to more fully address these sources of uncertainty and the potential impact on risk
estimates.
1.2 Purpose and Scope of Analysis
This document is a revision to the March TSD, and provides a revised description of the
national-scale risk assessment for Hg that was completed to inform the finding that it is
appropriate and necessary to regulate electric utility steam generating units in the United States
(U.S. EGUs), pursuant to Section 112(n)(l)(A) of the Clean Air Act (CAA). The appropriate and
necessary finding is based, in part, on an assessment of the potential public health hazard
-------
associated with Hg emitted from U.S. EGUs (see Section III of the preamble to the proposed
Mercury and Air Toxics Standard (MATS) for U.S. EGUs).5 Consequently, the risk assessment
was designed to assess whether a potential public health hazard is associated with Hg emitted
from U.S. EGUs. Because the focus of the analysis is to determine whether a potential public
health hazard exists and not to characterize the full range of risk associated with exposure to Hg
emitted from U.S. EGUs, this risk assessment can be viewed as a public health hazard screening
analysis. As such, the primary objective is to determine whether individuals exposed to Hg
emitted from U.S. EGUs through high-end consumption of freshwater self-caught fish have the
potential to experience significant risk. Conversely, the risk assessment is not intended to
represent a comprehensive assessment of risk for all segments of the population potentially
exposed to Hg emitted from U.S. EGUs. It is beyond the scope of this risk assessment to assess
the residual risk remaining after accounting for emissions reductions anticipated from MATS.
Given the purpose for the analysis, the following policy-relevant questions were
developed to guide the design of the risk assessment:
(a) What is the nature and magnitude the potential risk to public health in the
United States for individuals experiencing reasonable high-end exposure to Hg
in freshwater, self-caught fish that is attributable to current U.S. EGU Hg
emissions?
(b) What is the nature and magnitude of the potential risk for this same group of
individuals based on projected U.S. EGU Hg emissions in 2016 considering
potential reductions in EGU Hg emissions attributable to CAA requirements?6
and
(c) How is total risk from Hg exposure as estimated for both the current and future
scenarios apportioned between the U.S. EGUs and other sources of Hg? The
last policy-relevant question reflects the fact that Hg emitted from U.S. EGUs
does not result in a distinct and isolated exposure pathway, but rather is
combined with Hg emitted from other sources (domestic and international) in
contaminating fish. Therefore, to consider U.S. EGU contributions to exposure
and risk associated with the consumption offish containing Hg, we determine
what share of total Hg exposure is attributable to U.S. EGUs.
In addition to the above policy-relevant questions, the overall design and scope of the risk
assessment reflects consideration of a number of important technical factors related to air-
sourced Hg, including, in particular, Hg released from U.S. EGUs:
• While Hg exposure and risk can occur through a variety of pathways, the dominant
human exposure pathway associated with ambient air emissions is through the
consumption offish that have bioaccumulated Hg originally deposited to watersheds
following atmospheric release and transport. Deposition of Hg to watersheds includes Hg
originating from local/regional sources, combined with Hg that has been transported
5 76 FR 24976
6 For purposes of this analysis, we focus on 2016 as this is the year when compliance with mercury standards would
be required to occur.
-------
regionally and globally from global anthropogenic and natural sources. Generally
oxidized (divalent) and particle-bound Hg will deposit relatively closer to the release
source, while elemental Hg will travel much further, often becoming part of the global
pool, before being deposited.7
The scientific literature on mercury health effects provides the strongest support for
quantifying neurological deficits in children who were exposed to MeHg in utero through
maternal fish consumption.8
Mercury emitted by U.S. EGUs is likely to make a small contribution to MeHg in
foreign-sourced commercial fish consumed in the U.S. and in commercial fish caught
further off the U.S. coast. Therefore, this risk assessment, while acknowledging these
sources of exposure to U.S. EGU-sourced Hg, does not quantify these risks since the U.S.
EGU-attributable portion of these risks is likely to be small and any quantitative
estimates of U.S. EGU-attributable risk would be highly uncertain.9
While areas closer to the U.S. coast (including estuarine areas) as well as the Great Lakes
may have elevated U.S. EGU deposition in some cases, because of uncertainty in
modeling the linkage between U.S. EGU-attributable deposition and Hg in fish, we have
not included this commercial consumption pathway in the quantitative risk assessment.10
The type offish consumption likely to result in exposures to Hg with the greatest
contributions from U.S. EGU Hg emissions is associated with fishing activity at inland
freshwater rivers and lakes located in regions experiencing relatively elevated U.S. EGU
Hg deposition. Some watersheds had U.S. EGU contributions ranging up to 11% and
7 Mercury is a persistent, bioaccumulative toxic metal that is emitted from power plants in three forms: Gaseous
elemental Hg (Hg°), oxidized Hg compounds (Hg+2), and particle-bound Hg (HgP). Elemental Hg does not quickly
deposit or chemically react in the atmosphere, resulting in residence times that are long enough to contribute to
global scale deposition. Hg(2+) and Hg(p) deposit quickly from the atmosphere impacting local and regional areas
in proximity to sources.
8 The EPA's health benchmark for methylmercury exposure (the reference dose or RfD) is based on three
epidemiological studies. These studies relate hair mercury levels in mothers (a surrogate for exposure in utero) or
mercury in cord blood (a direct measure of fetal exposure) to deficits in children's performance on a range of neuro-
cognitive tests (see section 1.4.5).
9 While mercury emitted from U.S. EGUs does contribute to contamination of foreign-sourced commercial fish, the
fraction contributed by U.S. EGUs is small. Current estimates of U.S. EGU mercury emissions are -29 tons per year
(see section 2.3), compared with global anthropogenic mercury emissions (for 2005), excluding biomass burning,
estimated at approximately 2,320 tons (Pirrone et al., 2010; UNEP, 2010). Since the mercury in commercially
caught foreign-sourced fish also includes contributions by natural sources and re-emitted natural and anthropogenic
sources, estimated to be as high as 5,207 metric tons/year (Pironne et al, 2010; UNEP, 2010), the fraction
contributed by U.S. EGUs is quite small. Therefore, particularly in the context of estimating individual risk, U.S.
EGU contributions to risk that residents in the US experience through consumption of commercially caught foreign-
sourced fish, at present, is expected to be too small to characterize, given the uncertainties in determining the
fraction of mercury that is from U.S. EGUs. This observation would also likely hold for the U.S. EGU contribution
to commercial fish sourced from further off the U.S. coast, where total mercury loading is likely to also be
dominated by non-U.S. anthropogenic emissions which are globally transported.
10 While air quality modeling does suggest that some near coastal areas (e.g. the Chesapeake Bay) and portions of
the Great Lakes may have elevated U.S. EGU deposition relative to the average levels in the continental U.S.,
several factors make modeling how changes in mercury deposition affect fish tissue Hg concentrations in these near-
coastal areas and the Great Lakes challenging and uncertain. Specifically, the size of these waterbodies relative to
inland lakes and rivers and the potential for fish to have larger habitats makes it difficult to quantify the EGU
contribution to fish tissue Hg concentrations in these locations. Due to the greater uncertainty associated with
modeling near coastal watersheds, we have elected not to simulate this pathway in the risk assessment. Because we
have not assessed this pathway, this risk assessment may underestimate the number of at-risk watersheds in the U.S.
7
-------
higher in 2016 (see Section 2.4). Therefore, efforts to identify areas with high U.S. EGU-
attributable MeHg exposures and risk are focused on assessing risk for those areas that
have (a) relatively elevated fish tissue Hg concentrations and (b) relatively elevated levels
of U.S.EGUHg deposition.
Reflecting consideration of the policy questions and technical factors discussed above,
the scope of the national-scale Hg risk assessment is as follows.
• Hg deposition was modeled for two years, 2005 and 2016: The 2005 deposition is used
to represent deposition that is reflected in the sampled fish tissue MeHg levels for 2000 -
2009. The projected 2016 deposition reflects emissions after imposition of CAA
requirements. The available emissions data (see Section 2.3) suggest that current 2010
U.S. EGU emissions are closer to levels reflected in the projected 2016 Scenario and
substantially lower than levels reflected in the 2005 Scenario. As a result, the 2016
Scenario analysis is most relevant for this rulemaking. Further modeling of future
emissions indicates that in the absence of binding federal regulations U.S. EGU
emissions are not likely to be substantially reduced between 2010 and 2016 (although the
CAA requires the Agency to consider only federal CAA requirements in estimating
future HAP emissions and attendant risks associated with EGUs). The 2016 Scenario is
thus used to estimate risks related to both current emissions and emissions after
implementation of the CAA requirements and risks are not generated for the 2005
scenario.11 The estimated Hg emissions for U.S. EGUs in 2016, used in the deposition
modeling for this TSD, is 29 tons.
• Focus on assessing risk for female subsistence fish consumers of self-caught fish obtained
from inland freshwater watersheds: Given the goal of determining whether a public
health hazard is associated with U.S. EGU emissions, we have assessed risk for those
individuals likely to experience the greatest exposure and risk from consuming fish
impacted by Hg emitted from U.S. EGUs. This translates into a focus on women of child-
bearing age who consume subsistence-levels offish caught from inland freshwater
waterbodies.12 By focusing on inland waterbodies, we are focusing on those locations
with the greatest U.S. EGU-attributable Hg deposition (see Figure 2-4) and consequently
the greatest U.S. EGU-attributable fish tissue Hg concentrations. By focusing on
subsistence-level fish consumption scenarios, we focus on those self-caught fish
consumers with the highest intake rates and therefore, those who will experience the
greatest Hg exposures at a given watershed. In defining the high-end fisher populations
to include in the analysis, we have used peer-reviewed study data to characterize
consumption rates for a variety of high-consuming fisher populations that reflect different
SES groups and are active in different regions of the country (e.g., Laotians, Great Lakes
Tribal populations, Black and White anglers active in the Southeast - see Section 1.4.3).
As noted earlier in section 1.1 for the analysis, we focus on women of child-bearing age
11 Deposition estimates for the 2005 scenario are used in scaling fish tissue Hg concentrations to represent future
fish tissue levels in 2016, which are then used in turn to model total risk for the 2016 scenario (see section 1.4.6).
12 While subsistence-level consumption can reflect consumption by individuals whose SES status compels them to
supplement their diet with self-caught fish, elevated levels of inland freshwater self-caught fish consumption can
also be experienced by recreational anglers who fish often and consume a large amount of the fish they catch.
-------
since the MeHg RfD used in assessing risk is based on neurodevelopmental effects in
children exposed to mercury in utero.
Watershed-level assessment based on modeling female subsistence fish consumer risk at
watersheds where there is the potential for this type of high fish-consuming activity and
where we have fish tissue Hgdata: The risk assessment is conducted at the watershed
level, focusing on the subset of inland watersheds in the U.S. where we have fish tissue
Hg data and where we believe high fish-consuming subsistence fisher populations could
be active.13 Specifically, we generate a set of risk estimates for each watershed reflecting
the type of subsistence fisher activity that could potentially exist at that watershed.
Because it is not possible to enumerate these high-end fisher populations, the risk
estimates that are generated are not population-weighted and instead are given a uniform
weight for each watershed-level risk estimate generated.14 This focus on subsistence
fishing activity and associated consumption reflects the fact that this is a screening-level
analysis for risk associated with mercury emitted from U.S. EGUs and is not intended to
provide a comprehensive picture of the distribution of risk across all types of fishers
potentially active at watersheds where we have Hg fish tissue data.
Exclude commercial fish consumption from the quantitative risk analysis. Although risk
associated with commercial fish consumption may be a potential public health concern
under certain circumstances, the relatively low contribution of U.S. EGU Hg to this
source of dietary fish (relative to non-US Hg emissions), leads us to exclude this
consumption pathway from the risk assessment and to focus instead, on high-end self-
caught fish consumers. In the specific case of commercial fish sourced from near the U.S.
coast (e.g. Chesapeake Bay) and the Great Lakes, while there is the potential for U.S.
EGUs to have a greater role in affecting Hg levels in these fish, as noted earlier,
uncertainty associated with modeling the linkage between U.S. EGU Hg deposition and
Hg exposure and risk for this dietary pathway precludes us from including this pathway
in the risk assessment. We do not expect that exclusion of these scenarios will result in
significant downward bias in our overall estimates of the proportion of watersheds with
populations potentially at-risk due to U.S. EGU emissions of Hg, although the absolute
number of watersheds will be understated.
13 With the exception of the typical female subsistence fish consumer scenario (which is assessed across all
watersheds with fish tissue Hg data), the potential for the other SES-differentiated female subsistence fish consumer
scenarios to be active at a given watersheds is based on determining whether a group of similar SES-differentiated
individuals (referred to as a source population) lives in the vicinity of the watershed (see section 1.4.3).
14 In order to enumerate risk estimates generated for the female high-end consumer scenario used in this risk
assessment, we would need to have the following types of specific information: (a) the fraction of anglers who
consume at the subsistence-levels modeled for this population specifically at inland freshwater waterbodies, (b) for
this population, the fraction that focus their activity at individual watersheds, and target somewhat larger fish to
supplement their diet, and (c) for this subgroup, the fraction of female consumers of childbearing age who either fish
themselves and consume at this level, or obtain and regularly consume fish provided by fishers who focus their
fishing efforts at the individual watershed.. However, currently available information does not allow us to estimate
each of these subgroups of high-consuming fishers. Specifically, while we have data on the frequency of
recreational angling within the U.S., this covers general recreational fishing and not subsistence fishing. In this
analysis, we have focused on a subset of female subsistence-level fish consumers that we believe (a) could
potentially exist at a subset of watersheds evaluated in this analysis and (b) are likely to experience higher risk due
to their behavior (i.e., favor larger fish as a dietary source, likely to consume fish obtained primarily from individual
watersheds, consume larger amounts of self-caught fish). While we believe it is reasonable to assume that a subset
of high-end fishers would have these attributes, it is not possible at this point to definitively state at which
waterbodies they are active or to enumerate them for purposes of generating population-weighted risk distributions.
-------
• Include estimates of total risk from all Hg deposition sources, as well as the U.S. EGU
incremental contribution to total risk. As discussed below (Section 1.2), we focus on two
aspects of MeHg-related risk: (a) total Hg risk, with an estimate of the percent of that
total risk contributed by U.S. EGUs (i.e., the fraction of total risk associated with U.S.
EGUs) and (b) risk when deposition from U.S. EGUs is considered alone, before taking
into account deposition and exposures resulting from other sources of Hg. These two risk
metrics reflect the cumulative burden of Hg exposures and incremental contribution that
the U.S. EGU-attributable deposition makes to overall exposures to MeHg.15
1.3 Overview of Risk Metrics and the Risk Characterization Framework
The risk assessment generates hazard quotient (HQ) estimates by comparing estimates of
modeled potential exposure for subsistence fisher populations to the MeHg RfD. In addition to
the HQ estimates, Appendix A provides estimates of IQ loss in children born to mothers from
these high fish-consuming subsistence fishing populations. Because of concerns expressed both
by the SAB and by EPA staff that IQ loss may not fully capture the full range of
neurodevelopmental deficits in children exposed to MeHg in utero, we focus on the HQ
estimates in making our determination of the public health hazard associated with U.S. EGU Hg
emissions.16
As mentioned above in section 1.2 and discussed in greater detail in sections 1.4 and
1.4.6, we generate risk estimates at the watershed-level for the subset of watersheds in the U.S.
where we have Hg fish tissue data and where we believe the potential exists for high-end fish
consumption due to the presence of subsistence fishers. As noted earlier, limitations in
quantifying the number of high-consuming fishers active across the set of modeled watersheds
prevents us from generating population-weighted risk distributions. However, we do use the
watershed-level risk estimates (with uniform weighting across watersheds) to generate risk
15 When exposures are to be compared to the EPA's reference dose (RfD) for MeHg in order to generate a hazard
quotient (HQ), we must first consider total MeHg exposure given the definition of the RfD, which is intended to be
compared against total exposure to a given hazardous air pollutant. Once an HQ reflecting total exposure is
calculated, we can then consider the U.S. EGU incremental contribution to that total risk. However, U.S. EGU
incremental risk in the form of an HQ should not be considered in isolation without considering total risk associated
with MeHg in consumed fish.
16 Concerns have been raised in the literature that if mercury affects a set of specific neurological functions, then use
of full-scale IQ as the modeled health endpoint, could underestimate the neurodevelopmental impacts on other
targeted functions (Axelrad et al., 2007). In addition, two of the most sensitive endpoints in the Faroe Islands study
were the Boston Naming Test and California Verbal Learning Test, both of which can represent a significant
educational risk depending on severity, and those tests are not directly assessed as part of measuring IQ in children.
In addition, IQ does not cover other neurologic domains such as motor skills and attention/behavior and therefore,
risk estimates based on IQ will not cover these additional endpoints and could further underestimate overall
neurodevelopmental impacts (Axelrad et al., 2007). The wide range of neuropsychological effects potentially
associated with Hg exposure has also been highlighted by Grandjean et al., (1997) who described developmental
delays in verbal skills, learning and short-term memory, and more recently by Wijngaarden et al., (2006), who
provided benchmark dose calculations for 26 endpoints including a number of neuropsychological measures. These
studies highlight the range of neurodevelopmental effects in children potentially associated with Hg exposure. In
contrast to the IQ loss metric, the HQ metric based on the RfD reflects consideration for a wider array of
neurodevelopmental effects in children (e.g., Boston Naming Test, Continuous Performance Test, California Verbal
Learning Test, McCarthy Perceived Performance, McCarthy Motor Test, finger tap - U.S. EPA's Integrated Risk
Information System in 2001 - http://www.epa.gov/iris/subst/0073.htm).
10
-------
distributions from which we can identify specific risk percentiles, (e.g. the 95th percentile of risk
across watersheds). In this context, the distribution of modeled subsistence fisher risk estimates
represents the range of risk across modeled watersheds based on assessing the same subsistence
fisher scenario at each watershed (without population-weighting). Percentile risk estimates
obtained from this distribution are for the population of watersheds, rather than the population of
subsistence fishers.
Reflecting the goals of the analysis presented in Section 1.2, the HQ risk metrics are
calculated for both total risk and U.S. EGU-attributable risk. In considering U.S. EGU-
attributable risk, we generate two types of risk estimates:
• The percent or fraction of total risk attributable to U.S. EGUs at watersheds where total
risk is considered to pose a potential public health hazard: We consider the magnitude of
total HQ risk, identifying those watersheds with total MeHg exposure exceeding the RfD
(i.e., an HQ>1) and then estimate the fraction (or percent) of that total risk that is
attributable to U.S. EGUs.
• Risk focusing on U.S. EGUdeposition and excluding other non-U.S. EGU sources: For
this metric, we estimate risk based on the U.S. EGU incremental contribution to total
exposure. Specifically, for HQ, we compare U.S. EGU-attributable exposure against the
MeHg RfD.
In assessing the potential public health significance of HQ estimates, we considered total
MeHg exposures above the RfD to represent a potential public health hazard.17
Risk Characterization Framework
We have developed a 2-stage framework for using the risk metrics described above to
address the policy-relevant questions outlined in section 1.2. This 2-stage framework is
illustrated in Figure 1-1 and each of the stages is also described below (throughout this
document, this will be referred to as the "risk characterization framework"):
Stage 1 - Identify watersheds with populations potentially at-risk due to U.S. EGUHg based
on application of the two risk metrics:
a) The percent or fraction of total risk attributable to U.S. EGUs at watersheds
where total risk is considered to pose a potential public health hazard: Here we
identify watersheds with populations potentially at-risk due to U.S. EGU Hg by
identifying (a) those watersheds where total risk meets or exceeds levels
17 EPA's interpretation for this assessment is that any exposures to MeHg above the RfD are of concern given the
nature of the data available for Hg that is not available for many other chemicals, where exposures have often had to
be significantly above the RfD before they might be considered as causing a hazard to public health. The scientific
basis for the mercury RfD includes extensive human data and extensive data on sensitive subpopulations including
children exposed in utero ; therefore, the RfD does not include extrapolations from animals to humans, and for
database deficiencies. In addition, there is no evidence for a biological threshold observed for critical effect of
neurological deficits in children studied in the principal studies of the IRIS assessment for MeHg. This additional
confidence in the basis for the RfD suggests that all potential exposures above the RfD can be interpreted with more
confidence as representing a hazard to public health.
11
-------
considered to represent a potential public health hazard (i.e., HQ > 1); and (b)
U.S. EGUs contribute to total risk at this subset of watersheds with elevated risk
(we have considered various increments of U.S. EGU contribution ranging from 5
to 15%). Any contribution of Hg emissions from U.S. EGUs to watersheds where
potential exposures from total Hg deposition exceed the RfD is a hazard to public
health, but for purposes of our analyses we evaluated only those watersheds
where we determined U.S. EGUs contributed 5 percent or more to Hg deposition
in the watershed. EPA believes this is a conservative approach given the
increasing risks associated with incremental exposures above the RfD.
b) Risk focusing on U.S. EGU deposition and excluding other non-U.S. EGU
sources:: Here we identify watersheds with populations potentially at-risk due to
U.S. EGU-attributable risk (prior to considering Hg contributed by other sources).
Although this metric focuses on U.S. EGU exposure, it is important to keep this
incremental exposure in perspective with regard to total MeHg exposure in which
non-U.S. EGU sources of deposition typically dominate the U.S. EGU increment
across watersheds.
Stage 2 - calculate the combined (total) number of watersheds and percentage of watersheds
where populations may be at-risk from U.S. EGU-attributable Hg: Here we combine
estimates from Stages la and Ib to consider watersheds where populations may be at-
risk due to (a) U.S. EGUs contributing to exposures at watersheds where total risk
potentially poses a potential public health hazard or (b) U.S. EGUs making an
incremental contribution to total Hg exposure which, when considered separate from
total Hg exposure, represents a potential public health hazard.
This framework allows us to identify watersheds where U.S. EGU-related exposure
considered separately, or as a portion of total risk, represents a potential public health hazard.
More specifically, it allows us to estimate the number and percentage of watersheds where
populations may be at-risk due to U.S. EGU-related Hg emissions.
12
-------
Stage la: watersheds where total
MeHg exposure exceeds the MeHg
RfD and where US EGUs make a
substantial contribution (e.g., 5%) to
that exposure.
Stage lb: watersheds where U.S.
ECU deposition, if considered in
isolation, could result in MeHg
exposure that exceeds the MeHg RfD.
Some watersheds are identified in
both stages (these are not double-
counted in the totalStage2
number).
Stage 2: total numberof watersheds with
populations potentially at-risk due to Stage la
orStage lb analyses (i.e., union of Aand B) :
Stage2 = A + B-C
Figure 1-1. 2-Stage Risk Characterization Framework
1.4 Overview of Analytical Approach
This section provides an overview of the analytical approach for this risk assessment,
which we illustrate in Figure 1-2. In this figure, we identify the subsections of the TSD with
more detailed technical information on each of the analytical steps. We also provide a detailed
example calculation of the total HQ and U.S. EGU incremental HQ at the watershed-level in
section 1.6.
13
-------
Exposure modeling
Defining female subsistence consumer scenarios to
include in the analysis (section 1.4.3)
Identify which watersheds
will be modeledforeach
of the SES-differentiated
female subsistence
consumer scenarios*
Studies
characterizing fish
consumption rates
forfreshwater
subsistence fisher
populations
Characterizefemale subsistence
consumerscenariosto model
Characterizing
measured fish
tissue Hg
concentrations at
the watershed level
(section 1.4.2)
7
Estimating total fish consumption-
related MeHg exposure for fishers/
consumers active at each watershed
(section 1.4.4)
Projectfish tissue Hg
concentrations for the 2016
scenario at each watershed
(section 1.4.6)
Risk modeling
Estimate risk (RfD-based
hazard quotient) at each
watershed forthe 2016 scenario
(section 1.4.5)
Estimation of U.S. EGU-
attributable risk at each watershed
forthe 2016 scenario
(section 1.4.6)
CMAQ mercury
deposition modeling for
total and US EGU Hg
deposition ateach
watershed forthe 2005
and2016airquality
scenarios
(section 1.4.6.1)
* A more generalized typical female subsistence consumer scenario is also included in the analysis, which is
evaluated at all watersheds with fish tissue Hg data without consideration for the presence of a source population .
Due to its bro ad er g eographic coverage, th is scenario receives th e g reatest fo cus i n p resenti ng risk esti mates.
KEY:
decision
data input
analysis step
Figure 1-2 Flow Diagram of Risk Analysis (for the 2016 air quality scenario) Including
Major Analytical Steps and Associated Modeling Elements
The risk assessment assesses risk to female subsistence fish consumers from potential
exposure to MeHg from consuming fish caught in U.S. watersheds where we have measured fish
tissue Hg concentration data and where we have determined that subsistence fishing activity
could occur. As noted earlier in section 1.2, risks are only estimated for the 2016 scenario since
14
-------
underlying U.S. EGU emissions used in modeling this scenario are closer to actual 2010
emissions than are emissions estimates used in modeling the 2005 scenario. However, as
discussed below, Hg deposition estimates for the 2005 scenario are used in scaling fish tissue Hg
levels to represent future levels associated with the 2016 scenario (which are in turn used in
modeling risk for the 2016 scenario). We provide a brief description of each analytical step of
the risk analysis reflected in Figure 1-2:
11. Model mercury deposition: We modeled total and U.S. EGU-attributable mercury
deposition for the continental U.S. for 2005 and 2016 using the CMAQ model at 12 km
grid resolution. (See section 1.4.6.3).
12. Specify the spatial scale of the watersheds: We selected Hydrologic Unit Code (HUC) 12
as the appropriate spatial scale for watersheds, which tend to measure a few kilometers
(km) on a side and match up with the spatial resolution of the mercury deposition
modeling. We interpolated the gridded mercury deposition estimates to deposition for
each watershed in the continental U.S. (See section 1.4.1)
13. Characterize measured fish tissue Hg concentrations (75th percent! le) at each watershed.
We estimate the 75th percentile fish tissue Hg concentrations based on measurement data
collected primarily by the states collected from 2000-2010 for inland freshwater fish
species larger than 7 inches. We excluded watersheds without fish tissue data from
remainder of the analysis. (See section 1.4.2). The fish tissue data did not include
saltwater or estuarine fish.
14. Project 75th percentile fish tissue Hg concentrations for the 2016 scenario: We used the
assumption of a proportional relationship between Hg deposition and fish tissue Hg
concentrations together with deposition estimates for the 2005 and 2016 scenarios to
project fish tissue Hg concentrations for the 2016 scenario. Specifically, we used the ratio
of 2016 deposition to 2005 deposition at a given watershed to adjust the 75th percentile
fish tissue Hg concentration discussed in Step 3, to represent a 2016 concentration for
that watershed (see section 1.4.2.1).
15. Define female subsistence fish consumer scenarios: The analysis focused on females who
consume subsistence-levels offish that were caught at inland freshwater waterbodies.
Our literature review identified seven female subsistence fish consumer scenarios that we
selected to include in the analysis: (1) typical female subsistence fish consumer, (2) low
income White fishers in the southeast, (3) low income Black fishers in the southeast, (4)
low income Hispanic fishers, (5) Vietnamese fishers, (6) Laotian fishers, and (7) Tribal
fishers (Chippewa active near the Great Lakes). (See section 1.4.3).
16. Identify watersheds with subsistence fisher population activity: For the typical female
subsistence fish consumer scenario, we assume subsistence fisher population activity at
all watersheds where we have fish tissue Hg data. For the Tribal fishers, we assume
activity within all watersheds with fish tissue Hg data located within territories ceded to
these tribes. For the remaining scenarios, we use demographic data to determine if at least
25 individuals with SES attributes matching those of the subsistence consumer
population are present in the vicinity of a watershed, and if so, we assume that
subsistence fishing activity might occur at that watershed. (See section 1.4.3)
15
-------
17. Define self-caught fish consumption rates for the subsistence scenarios: We used survey
data published in the literature to identify self-caught fish consumption rates for each
female subsistence fish consumer scenario. In each case, at the high end (e.g., 90th to 99th
percentiles), consumption rates reached subsistence levels, ranging from an 8oz self-
caught fish meal every few days to an 8oz meal every day and higher in some cases. (See
section 1.4.3).
18. Estimate total fish consumption-related MeHg exposure (2016 scenario): We estimated
exposure in the form of daily-average MeHg intake at the watershed-level for each
female subsistence fish consumer scenario potentially active at a given watershed. These
estimates of exposure use the projected 75th percentile fish tissue Hg concentrations for
2016 described in Step 4 above. We were careful to match the fish sampling data with
type of sampling (e.g., filet skin on, whole fish), cooking adjustment, and type of
consumption rates (e.g., as purchased, as consumed). The exposure estimates represent
potential exposure to MeHg from fish caught in the watershed, not population-weighted
exposures. (See section 1.4.4)
19. Estimate of total MeHg risk (RfD-based HQ) at each watershed (2016 scenario): We
compare the watershed-level exposure estimates modeled for each female subsistence
fish consumer scenario (for 2016) to the MeHg RfD to generate HQ estimates for total
MeHg exposure. (See section 1.4.5)
20. Estimate of U.S. EGU-attributable risk (2016 scenario): We used the same
proportionality assumption discussed in Step 4, together with total and U.S EGU-
attributable Hg deposition estimates (generated at the watershed-level) for the 2016
scenario to estimate the fraction of total HQ (for the 2016 scenario) attributable to U.S.
EGUs. Specifically, we used the ratio of U.S. EGU-related deposition to total deposition
(for the 2016 scenario) at a given watershed calculate the fraction of total risk (generated
in Step 9) that is attributable to U.S. EGUs (see section 1.4.6).18
1.4.1 Specifying the spatial scale of watersheds
The first step in designing the analysis was to specify the spatial scale of the watersheds
to use as the basis for risk characterization. As noted above, this risk assessment is based on
estimating risk at watersheds for which we have measured Hg fish tissue data. Two studies
(Knights et al., 2009, Harris et al., 2007) examining the response of aquatic freshwater
ecosystems to changes in Hg deposition focused on watersheds with dimensions closest to HUC-
12. In each of the studies, researchers used watersheds reflecting a fairly refined spatial scale
18 Given the assumption that fish tissue Hg concentrations and exposure and risk are linearly related, the
proportionality assumption (together with the ratio of U.S. EGU-related to total Hg deposition for the 2016 scenario)
could be applied either to the fish tissue Hg concentrations or to total risk in order to generate U.S. EGU-attributable
risk. For the analysis, as reflected in Step 10, we applied the proportionality assumption and deposition ratio once
total risk (for the 2016 scenario) had been generated. However, we could have used the proportionality assumption
to estimate the fraction of the 75th percentile fish tissue Hg concentration (projected for the 2016 scenario) that was
attributable to U.S. EGUs at each watershed and then used that fractional fish tissue value to estimate U.S. EGU-
attributable risk for the 2016 scenario. Both approaches would generate the same risk estimate, reflecting the
underlying linearities in the exposure and risk models.
16
-------
(approximately 5-10 km on a side). This suggests that, at least in the context of these studies,
researchers believed that the relationship between changes in Hg deposition and changes in
MeHg levels in aquatic biota could be effectively explored at the level of these more spatially
refined watersheds. Each of the studies is briefly summarized below.
The Knights et al., 2009 study focused on characterizing the temporal pattern of
reductions in fish tissue Hg concentrations following reductions in Hg deposition over
waterbodies and associated watersheds. This study relied on modeling and included simulation
of five different types of waterbodies ranging from a seepage lake (with little watershed loading)
in Florida to a stratified drainage lake in NH. The scale of the five watersheds included in the
Knights et al., 2009 study range from 20 by 100 km (for the coastal plain river location in GA) to
5 by 10km (for the Lake Waccamaw NC site). Three of the five locations had watersheds in the
10 by 10km range (see Figure 2 in the article). Given that the majority of locations in the study
had smaller watersheds (i.e., in the 10 by 10km range), we conclude that this would represent a
reasonable watershed spatial scale to use in linking changes in aerial deposition to changes in
fish tissue levels (i.e., as the basis for risk characterization in the analysis).
An article by Harris et al., 2007, which is based on the METALLICUS study (specifically
lake 658 catchment in northwestern Ontario, Canada), also examined the temporal profile
associated with changes in media and biota Hg levels following a change in Hg deposition. In
this study, a 3-yr loading of radio-labeled Hg to the waterbody and watershed (separate labeled
Hg applied to each location) was followed by measurement of Hg in various media and biota to
see how long it took for the loaded Hg to impact different media compartments. The single
watershed involved in this study is relatively small (only a few km on a side). Therefore, the
spatial scale of the watershed involved in this study also supports use of a more refined spatial
scale for watersheds in the risk assessment.
In addition to considering the scale of watershed reflected in these two studies of Hg
loading response, use of a more refined spatial scale (i.e., use of HUC12s rather than a coarser
scale of watershed) in linking changes in Hg deposition to changes in fish tissue Hg
concentrations also reduces the potential for averaging out areas of high Hg deposition. The
HUC12 represents the most refined scale of watershed currently available at the national level
and therefore was chosen as the basis for linking changes in Hg deposition to changes in fish
tissue Hg concentrations. Conversely, use of larger watersheds, while allowing us to model
more of the country in the risk assessment, could result in the unwarranted dilution of areas of
elevated Hg deposition from U.S. EGUs (and therefore, by association, the dilution of U.S.
EGU-attributable Hg exposure and risk). The SAB expressed support for the use of HUC12-scale
watersheds.
1.4.2 Characterizing measured fish tissue Hg concentrations at the watershed level
The next step in the analysis was to characterize fish tissue Hg concentrations at the
watershed level using measured fish tissue Hg data. This process involved three tasks: (a)
develop the database offish tissue Hg concentrations based primarily on state-level data, using a
number of filtering steps that are described below, (b) use the fish tissue database to generate
percentile estimates (75th and 50th percentiles) for each watershed containing fish tissue Hg
17
-------
measurement data and (c) filter the set of watersheds with fish tissue Hg percentile estimates to
exclude locations potentially impacted by non-atmospheric Hg sources.
The SAB recommended that EPA evaluate whether there were additional fish tissue data
that could be used to update the fish tissue dataset (see Section 1.1). We implemented this
recommendation, and as a result, the fish tissue Hg dataset used in the revised national-scale risk
assessment is an augmented version of that used in the March assessment. Because the fish
tissue Hg dataset is different from the March TSD, this section provides an overview of the 2010
Mercury Fish Tissue (MFT) dataset as well as the augmentation steps of the 2010 MFT dataset
based on additional fish tissue data identified for a handful of states. We have also included
additional detail on the derivation of the fish tissue Hg dataset in response to an SAB
recommendation. We describe each step to develop the dataset ultimately used in this risk
assessment, which is illustrated in Figures 1-3 through 1-6. We also provide the number offish
tissue Hg measurements and watershed-level estimates associated with each step in the filtering
process.
Development of the 2010 Mercury Fish Tissue (MFT) database used in the March version of the
national-scale mercury risk assessment
To develop the 2010 MRT dataset, we began with fish tissue samples from three main
sources:
• National Listing of Fish Advisory (NLFA) database. The NLFA, managed by EPA
(http://water.epa.gov/scitech/swguidance/fishshellfish/fishadvisories/), collects and
compiles fish tissue sample data from all 50 states and from tribes across the United
States. In particular, the 2010 version of the NLFA used in this analysis contains data for
over 45,000 Hg fish tissue samples collected from 1995 to 2007.
• U.S. Geologic Survey (USGS) compilation of mercury datasets. As part of its
Environmental Mercury Mapping and Analysis (EMMA) program, USGS compiled Hg
fish tissue sample data from a wide variety of sources (including the NLFA) and has
posted these data at http://emmma.usgs.gov/datasets.aspx. To avoid duplication in our
analysis, we excluded all of the USGS data originating from the NLFA. As shown in
Figure 1-3, we included data from the USGS compilation that originated from two main
categories of sources:
(1) state-agency collected and reported data (including Delaware, Iowa, Indiana,
Louisiana, Minnesota, Ohio, South Carolina, Virginia, Wisconsin, and West Virginia)
from nearly 40,000 fish tissue samples, covering the period 1995 to 2007 (referred to
as "USGS States" in Figure 1-3)
(2) over 6,000 fish tissue samples from several other sources, including the National Fish
Tissue Survey, the National Pesticide Monitoring Program (NPMP), the National
Contaminant Biomonitoring Program (NCBP), the Biomonitoring of Environmental
Status and Trends (BEST) datasets of the USFWS and USGS
(http://www.cerc.cr.usgs.gov/data/data.htm), and the Environmental Monitoring and
18
-------
Analysis Program (EMAP) (http://www.epa.gov/emap/). (referred to as "USGS
Other" in Figure 1-3)
• EPA's National River and Stream Assessment (NRSA) study data. These data include
nearly 600 fish tissue Hg samples collected at randomly selected freshwater sites across
the United States during the period 2008 to 2009.
All of the measurements in the NLFA database were obtained from state fish advisory
programs managed by the health department, natural resource department or environmental
protection agency in each state. In many cases, these data were not sampled randomly but instead
reflect protocols intended to target (1) areas know to support recreational and or commercial
fishing and (2) areas believed to have elevated levels of chemical contamination (i.e., MeHg) in
fish.19 Therefore, the state-level data do not provide a representative characterization of the
distribution of MeHg concentrations in fish tissue at waterbodies across the state and are instead,
likely to be biased towards locations with higher Hg fish tissue concentrations as well as species
that are typically consumed by the general public. However, because the goal of this analysis is
to determine the potential for a public health hazard from Hg emitted by U.S. EGUs, it is
beneficial to have fish tissue measurement data potentially biased towards waterbodies with
greater Hg impacts as well as species typically consumed by recreational or commercial fishers
because this reduces the likelihood that high risk watersheds would be omitted due to gaps in fish
sampling.
The majority offish tissue Hg measurements in these datasets are for total Hg and not
MeHg. However, research published in the literature suggests that 90 to 95 (or greater) percent
of total Hg in fish tissue is MeHg (U.S. EPA, 2000). Based on this research, for purposes of
assessing exposure and risk, we have assumed that 95% of each fish tissue Hg concentration is
MeHg (see section 1.4.4).
Data from the four datasets shown in Figure 1-3 were combined into a single "2010
Master" Hg fish tissue (MFT) sample dataset covering the period 1995 to 2009. One problem
encountered in combining these datasets is the potential duplication of samples in the NLFA and
19 We reviewed approaches used by Pennsylvania and Wisconsin for determining where to collect fish tissue Hg
measurements. These states were chosen because they either have relatively high fish tissue Hg concentrations
(Wisconsin), or relatively elevated U.S. EGU-related Hg deposition (Pennsylvania). Pennsylvania identifies
sampling locations based on (a) a need to verify or delist specific locations for fish advisories, (b) provide additional
data points for their Water Quality Network and (c) respond to requests from selected State Environmental and
Wildlife officials to sample locations of potential concern (Pennsylvania Department of Environmental Protection's
"Fish Tissue Sampling and Assessment Protocol" available at:
http://www.portal.state.pa.us/portal/server.pt/communitv/fish consumption/10560 ). Wisconsin selects sampling
locations based on providing coverage for (a) northern parts of the state where high Hg fish tissue levels have been
found, (b) areas were pollutions impacts are suspected, (c) areas with high fishing pressure and (d) locations that are
considered "indicators" for specific watersheds (Wisconsin Department of Natural Resources, "A Summary of
Mercury Concentrations in Fish (edible portions) from Wisconsin Waters 1990-2005", available at:
http://dnr.wi.gov/fish/consumption/1990-2005mercurvsummarv.pdf). While the details of these two strategies do
differ somewhat, generally both states are targeting areas either suspected of having a potential Hg problem, or areas
with elevated fishing activity. In both cases, the set of fish tissue Hg concentrations are not going to be generally
representative of trends across watersheds in the state and instead, are likely to favor coverage of more heavily
impacted watersheds (at least for the fraction of samples collected to cover areas of potential concern with regard to
mercury impacts).
19
-------
USGS-States data. Unfortunately, these two datasets do not contain directly comparable and
unique identifiers that allow duplicate samples to be easily identified and removed. In order to
identify potentially duplicative samples, we subdivided the samples from these two datasets into
data groups according to the year and state in which they were collected. If both datasets
contained a data group for the same year and the same state, then the data group with the fewer
number of observations was not included in the master data. This process excluded 18,860
potentially duplicate samples from the master dataset.
In finalizing the 2010 MFT sample dataset, the following filters were used to further
screen the fish tissue samples:
(a) excluded samples that were not geo-referenced (i.e., did not include latitude and
longitude information)
(b) excluded samples with missing date information
(c) excluded samples that were not from freshwater fish species or freshwater locations
(i.e., excluded estuarine locations),
(d) excluded samples from fish smaller than 7 inches in length. 20 Additional discussion
of uncertainty related to this filtering step is presented later in this section.
These filters excluded a total of 21,674 samples from the dataset (in addition to the 18,860
duplicates referenced earlier). For the majority of measurements in the database, we also have
information on the type of waterbody (river/lake) and type of sampling method used (e.g., filet
skin on). Additional detail on the process used to develop the master fish tissue dataset can be
found in U.S. EPA, 201 la, Section 5.2.2.
Even though we compiled fish tissue Hg sampling data for 1995 to 2009, we decided to
restrict data to 2000 to 2009 in the risk assessment (the SAB supported this decision). We
excluded fish tissue samples that likely reflected Hg deposition levels from the 1990's when
anthropogenic emissions in the U.S. were higher than after 2000. We recognize the complex
spatial and temporal nature of the response offish tissue Hg concentrations to changes in Hg
deposition and loading and acknowledge that a portion of the sampling data from 2000 to 2009
could still reflect higher Hg loading rates from earlier periods. Excluding the samples collected
before 2000 further reduced the size of the initial dataset by 27,522 observations (the potential
impact on the risk assessment related to our decision to focus on data from 2000 and later is
discussed in section 2.7, Table 2-15, Entry I).
After completing all of these filtering steps, the 2010 MFT sample dataset contained
23,770 observations (samples) in the U.S.
20 Seven inches represents a minimum size limit for a number of key edible freshwater fish species established at the
State-level. For example, Pennsylvania establishes 7 inches as the minimum size limit for both trout and salmon
(other edible fish species such as bass, walleye and northern pike have higher minimum size limits (Summary Book:
2001 Pennsylvania Fishing Laws and Regulations available at:
http://fishandboat.com/fishpub/summary/inland.html').
20
-------
Original Data
Sources
(>=1995)
(45,166)
USGS States
(39,972)
USGS Others
(6,098)
Data Merging
Data Screening
Reason
Dropped
Merged NLFA-
USGS States
(66,278)
Missing
Lat/Long
Missing Date
Not a
Freshwater
Fish
Not a
Freshwater
Location
Length < 7
inches
Yea r< 2000
Total
-9,177
-1,520
-2,638
-3,192
-5,147
-27,522
-49,196
2010 MFT
Sample
Dataset
(23,770)
Figure 1-3 Diagram Illustrating Step-wise Procedure Used to Develop 2010 Mercury
Fish Tissue (MFT) Dataset Used in the 2010 National-Scale Mercury Risk
Assessment
Augmentation of the 2010 MFT database with additional fish tissue mercury data obtained for a
subset of states (i.e.. ML NJ. MN. PA. and WI).
Since the March TSD, we augmented the 2010 MFT sample database with additional
sampling data from selected states and years. The composition and processing of these
"augmentation" data are shown in Figure 1-4. All of these data were provided to EPA by the
individual states, but they have not yet been incorporated into the most recent versions of the
NLFA. The main criteria for selecting these particular 5 states are: (1) a majority of the sample
data they provided to EPA were geo-referenced and included the year in which they were
sampled (more recent data for a number of the other states did not have these critical descriptors)
and (2) the states are located in areas of the country with relatively high levels of Hg deposition.
To avoid potential duplication of samples with those included in the initial Hg fish tissue
database, the augmentation data only included samples for the years 2003 to 2009 for MI, NJ,
MN, and PA, and for 2000 to 2010 for WI (Figure 1-4 provides more detail).21 All of the data
were reorganized and converted into Microsoft Access databases, using the NLFA fields and
21 The version of the NLFA used in developing the 2010 MFT dataset did not include samples collected between
2003 and 2009 for MI, NJ, MN or PA. Similarly, that earlier version of the NLFA did not include any WI samples
collected between 2000 and 2010. Therefore, addition of samples from the latest version of the NLFA from these
years for these 5 states would supplement the MFT without duplication.
21
-------
formats. They were then combined into the "Augmentation Master" dataset shown in Figure 1-4,
which includes a total of 12,864 samples.
Using the same set of screening steps, we ended up with an Augmentation MFT sample
dataset containing 11,797 observations in the U.S.22
For this analysis, we then combined the 2010 MFT sample dataset with the Augmentation
MFT dataset (as shown in Figure 1-5) to create the 2011 MFT sample dataset. This combined
dataset includes 35,567 fish tissue Hg samples from years 2000-2010. As with the original 2010
MFT database, samples from the 2011 MFT database are located across the U.S.; however, they
are more heavily focused in locations east of the Mississippi River.
Original Data
Sources
Ml 2007-2009
NJ 2007
MN 2003-2007
(6,012)
PA 2003-2008
Wl 2000-2010
(6,042)
Data Merging
Augmentation
' Master
(12,864)
Data Screening
Reason
Dropped
Missing Lat/Long -643
Missing Date -0
Not a Freshwater -13
Fish
Not a Freshwater -0
Location
Length < 7 inches -411
Yea r< 2000 -0
Total -1,067
Augmentation
MFT Sample
Dataset
(11,797)
Figure 1-4 Diagram Illustrating Step-wise Procedure Used to Develop the Augmentation
Mercury Fish Tissue (MFT) Dataset
22 The data fish tissue sample data from PA did not include a measure of fish length; therefore, all of the samples
were retained and assumed to be above the 7 inch threshold. In the March assessment, these samples were excluded,
however, in order to provide a more complete spatial coverage in PA, which has high levels of U.S. EGU Hg
deposition, we elected for this revised assessment to include these samples.
22
-------
Fish tissue
samples
2010 MFT Sample
Dataset
(23,770)
Spatial
Aggregation
(associating individual fish
tissue sampling locations
with HUC12watersheds-
completed using GIS)
Augmentation MFT
Sample Dataset
(11,797)
2011 MFT Sample
Dataset
(35,567)
230 observations (associated with 15 HUC12
watersheds) dropped because watersheds contain
goldmines or have substantial non-EGU emissions*
Fish tissue
samples
associated
with specific
HUC12
watersheds
2010 MFT HUC-
Level Dataset*
(2,323 HUCs)
Augmentation
MFTHUC-
Level Dataset
(940 HUCs)**
2011 M FT H DC-Level
>ataset:
(3,147)
*An additional 11 observations were dropped due to 5 HUC locations outside of the continental U.S.
**116of these 940 HUCs were also includedin the2010MFTdataset.
Figure 1-5 Diagram Illustrating Step-wise Procedure Used to Combine the 2010 MFT and
Augmentation MFT Datasets
Use the fish tissue database to generate percentile estimates (75th and medians) for each HUC-12
watershed containing fish tissue Hg measurement data
To conduct the risk and exposure analysis, we then spatially aggregated the 2011 MFT
sample data to the watershed level. To begin this process, we used the latitude and longitude
information from each sampling location to identify the HUC-12 watersheds in which each site
was located.
As a final data screening step, we then excluded HUC-12 watersheds (and the samples
located within those watersheds) that either contained active gold mines or had other substantial
non-U.S. EGU anthropogenic emissions of Hg.23 These watersheds were excluded because the
assumption of linear proportionality between Hg deposition and fish tissue Hg concentrations,
supported by the MMaps study is most supportable in those situations where aerial deposition is
the dominant source of Hg loading to a watershed. We identified watersheds with gold mines
23 The filtering out of locations with potentially significant non-air mercury loadings was also completed in
generating the 2010 MFT database. At this stage, to restrict the analysis to the continental U.S., where we have Hg
deposition modeling, we also excluded 1 1 samples that were located in 5 HUC-12 watersheds in Alaska.
23
-------
using a USGS data set characterizing mineral and metal operations in the U.S. (USGS, 2005).
The data represent commodities monitored by the National Minerals Information Center of the
USGS, and the operations included are those considered active in 2003. We identified
watersheds with substantial non-EGU anthropogenic emissions using a TRI-net query for 2008
for non-EGU Hg sources with total annual on-site Hg emissions (all media) of 39.7 pounds or
more.24 This threshold value corresponds to the 25th percentile annual U.S.-EGU Hg emission
value as characterized in the 2005 NATA.25 The 25th percentile U.S.-EGU emission level was
selected as a reasonable screen for additional substantial non-U.S. EGU emissions to a given
watershed. The SAB endorsed this as a sound approach, while noting that the degree of
conservatism implied by this approach is unknown. The SAB also commented that while other
screening criteria could be applied, such as removing watersheds near urban areas with potential
waste runoff, these would be unlikely to substantially change the results. In addition, applying
additional screening criteria would have the negative impact of further reducing the geographic
scope of the analysis.
Application of the filtering described here (i.e., excluding locations with active gold
mines or other substantial non-U.S. EGU anthropogenic emissions of Hg) resulted in 15 HUC12
watersheds, containing a total of 230 observations, being excluded from the 2011 MFT dataset
(see Figure 1-5). The final number of HUC12s with fish tissue Hg data included in the risk
assessment is 3,141 (containing 35,567 samples), which represents a 33% increase in the number
of watersheds assessed since the March TSD, as summarized in Figure 1-5.
As shown in Figure 1-5, this approach yielded 3,141 HUC-level sets offish tissue Hg
estimates for the augmented (full) dataset used in the revised national-scale mercury risk
assessment (this compares with 2,317 HUC-level fish tissue Hg estimates used in the March
version of the risk assessment).
Because most HUC-12 watersheds with measured fish tissue Hg data have multiple
sampled values, often distributed over multiple sampling sites, we needed to identify and
calculate summary statistics for each watershed, in order to represent fish tissue Hg levels in
estimating exposure and risk. To do this, we generated summary statistics (means and
percentiles) for MeHg concentrations in each HUC-12 and compiled these summary data into a
separate HUC-level dataset. The following two-step procedure was used to generate HUC-level
statistics: (a) calculate mean and percentile (25th, 50th, 75th and 90th) fish tissue Hg values for
each sampling site within a HUC and (b) for the mean, or a given percentile, take the average of
the applicable values across the sampling sites within a HUC to generate a single HUC-level
estimates (for the mean, or specific percentile). For example, if we had two sampling sites (each
with multiple samples) and we wanted a 75th percentile fish tissue Hg value for that HUC, we
would first compute the 75th percentile values at each sampling site and then take the average of
those two 75th percentile values. If there is only one sampling site in a given HUC, then we
simply compute the mean or percentile of interest from the measurements at that site and that
statistic is used to represent the HUC. This approach reflects a fisher who fairly consistently
targets the same size (represented by percentile) fish at each location where they fish within a
24 U.S. EPA Toxics Release Inventory (TRI) accessed at: http://www.epa.gov/tri/
25 U.S. EPA 2005 National Air Toxics Assessment (2005 NATA) accessed at:
http://www.epa.gov/ttn/atw/nata2005/..
24
-------
watershed. We are then constructing a representative fish tissue Hg value for them, by taking the
average of the fish they are simulated to catch across their fishing locations (within a given
HUC). As noted earlier, our simulation of female subsistence fish consumer exposure and risk
reflects the assumption that fishing activity is targeted within a given watershed.26
Summary statistics for the 2011 MFT HUC-level datasets are reported in Table l-l.27 The
average number offish tissue measurements for the period 2000 to 2009 for the 3,141
watersheds is 11.23, although some watersheds contained up to 360 measurements. Most
watersheds also have multiple species (mean of 2.69) and to a lesser extent, multiple sampling
sites within the watershed (mean of 1.41). Multiple sampling sites in this context can be different
streams and/or lakes located within the same watershed. Distributions of other HUC-level
percentiles (and the mean) are also presented for purposes of comparison.
Table 1-1 Summary Statistics for the 2011 MFT HUC-level Data
Distribution of various attributes across the 3,141 HUCs
Number of samples per HUC
Number of species per HUC
Number of sampling sites per HUC
Average of different HUC-level Hg concentration
percentiles (including the mean) across the 3,141 HUCs
Average of location-specific mean Hg concentrations (ppm)
in HUC
Average of location-specific 25th percentile Hg
concentrations (ppm) in HUC
Average of location-specific 50th percentile Hg
concentrations (ppm) in HUC
Average of location-specific 75th percentile Hg
concentrations (ppm) in HUC *
Average of location-specific 90th percentile Hg
concentrations (ppm) in HUC
Mean
11.23
2.69
1.41
Std.
Dev.
21.05
2.19
1.25
Min
1
1
1
Max
360
17
33
0.27
0.19
0.25
0.32
0.40
0.24
0.19
0.23
0.31
0.41
0
0
0
0
0
3.56
2.20
3.56
6.61
7.35
* 75th percentile row bolded since this is the statistic used in the risk assessment
26 The SAB recommended that a single fish tissue Hg value not be used to represent a watershed when there are
multiple measurement sites (e.g., multiple lakes) within that watershed. This recommendation reflected concern that
waterbodies, even when in close proximity within the same watershed, can display different methylation rates such
that they are likely to respond differently to a unit change in Hg deposition. The approach outlined here for
calculating the 75th percentile fish tissue Hg value does not use a single fish tissue Hg value in instances where there
are multiple sites within a watershed. Instead, we calculate the 75th percentile fish tissue Hg for each site (i.e., with
each value representing a specific sampling site, which often translates into a distinct waterbody) and then take the
average of those values, reflecting fishing activity that is distributed across the sampling sites, or waterbodies within
a specific watershed.
27 In the 2010 MFT dataset used in the March version of the risk assessment, approximately 48% of the fish tissue
Hg samples were obtained from rivers (with the rest coming from lakes). For the augmentation dataset, we did not
specify river versus lake in developing the fish tissue Hg dataset, since this distinction was downplayed by the SAB
as a factor significantly impacting application of the proportionality assumption in the context of estimating the U.S.
EGU-attributable portion of exposure and risk.
25
-------
-th
As noted in Section 1.4, we selected the 75 percentile fish tissue value at each watershed
28
th
as the main basis for exposure and risk characterization. Selection of the 75 percentile value
was based on the assumption that a subset of subsistence fishers would favor larger fish which
have the potential for higher bioaccumulation (i.e., use of a median or mean value could low-bias
likely catch-related Hg levels). There is uncertainty associated with this assumption and should
fishers at a particular watershed favor fish that are either larger or smaller than the type offish
reflected in the 75th percentile sample, risk estimates could be biased accordingly. Uncertainty
related to use of the 75th percentile fish tissue Hg value in exposure modeling (specifically,
potential bias in the estimation of the 75th percentile value) is discussed in section 2.9, Table 2-
15, Entry C. As recommended by the SAB, we also included of a sensitivity analysis using the
median fish tissue Hg concentration at each HUC - see section 2.8).
Number of HUC12 Locations
201OM FT Data
(#of HUCs)
2,201
Augmentation MFTData
(# of HUCs)
116
824
2011 MFT Data (Combining 2010 and Augmentation)
Total Number of HUC12 observations = 3,141
Figure 1-6 Diagram Illustrating Number of HUC12s with Fish Tissue Mercury Data (for
2010 MFT, Augmentation MFT and the Combined 2011 MFT Datasets)
Uncertainty related to low sample size for a substantial fraction of the HUC12s and exclusion of
fish <7 inches in length in calculation the HUC-level percentile estimates
While the majority of the 3,141 watersheds included in the risk assessment do have
multiple measurements, a substantial fraction (41%) have only 1-2 samples, which potentially
biases low a 75th percentile estimate for those watersheds. To examine the potential magnitude of
low bias in the 75th percentile estimates, we have summarized the distribution of 75th percentile
28 We note that the 75th percentile value is a constructed statistic that reflects the fish tissue samples available at a
given watershed. For some watersheds the number of samples is very low, 1 or 2 in some cases (see Table 1-2), and
as a result, estimation of the true 75th percentile of fish tissue mercury concentrations in those watersheds is likely to
be biased low. This is discussed later in this section. Depending on the degree to which the underlying fish tissue
data are representative of actual fish targeted by subsistence fishers active at that waterbody, this statistic may or
may not actually represent fish tissue Hg concentrations associated with fish consumption by those fishers.
26
-------
fish tissue values (as used in the risk assessment) for various sample-size-based strata of the
3,141 watersheds (see Table 1-2). We have also included two plots (as recommended in the
SAB peer review) including (a) a histogram comparing the frequency of different sample sizes
across the HUCs included in the risk assessment (Figure 1-7) and (b) a LOESS (locally-weighted
scatter plot smoothing)-based fitted curve relating the 75th percentile HUC-level values and the
frequency of samples sizes by HUC (Figure 1-8).29 The number of sampling sites also varies
across watersheds included in the risk assessment, with some watersheds having up to 30 or
more distinct sampling sites.
While supporting the use of the 75th percentile fish tissue Hg value in the risk assessment,
the SAB recommended a sensitivity analysis involving use of a median fish tissue Hg value in
place of the 75th percentile value. This sensitivity analysis is discussed in section 2.6. However,
we have provided fish tissue summary statistics here for median fish tissue statistics to support
comparison with the 75th percentile values (Table 1-3).
Table 1-2 Summary of 75th Percentile Hg Concentrations in Fish Tissue Samples by
Number of Observations and Number of Sites per HUC
Number of
Observations per HUC
1 Observation
2 Observations
3-5 Observations
6-10 Observations
> 10 Observations
Total
Number of sampling
sites per HUC
1 Site
2 Sites
3-5 Sites
6-10 Sites
> 10 Sites
Total
Percentiles (of the 75th percentile HUC-level fish tissue
Hg concentration - ppm)
25th
0.077
0.09995
0.12
0.185
0.25
0.1294
25th
0.12
0.160967
0.179823
0.125972
0.2698
0.1294
50th
0.1564
0.1775
0.205625
0.282
0.37
0.245
50th
0.233
0.276706
0.299482
0.224625
0.410415
0.245
75th
0.2887
0.315
0.339
0.464
0.553
0.422684
75th
0.4
0.45044
0.478662
0.479822
0.494276
0.422684
90th
0.493
0.5135
0.577
0.743
0.85
0.673355
90th
0.672
0.71
0.603667
0.693108
0.847259
0.673355
95th
0.67834
0.765
0.8
0.985
1.083412
0.893
95th
0.87
1.047125
0.738476
0.8652
0.847259
0.893
Mean
0.226803
0.256943
0.283682
0.374177
0.444645
0.324052
Mean
0.31308
0.360398
0.350198
0.325642
0.419069
0.324052
N
838
458
468
506
871
3141
N
2371
553
196
19
8
3147
Based on the observations-related data presented in Table 1-2 and Figure 1-8, we
acknowledge the potential for low-bias in the 75th percentile values for HUCs with lower sample
sizes. Specifically, in Table 1-2, the mean 75th percentile HUC-level value increases across
strata as the sample size increases. This suggests that as the number of samples increases,
estimates of the 75th percentile will be more likely to approach the true value, and thus are more
likely to represent higher end fish tissue levels. Similarly, in Figure 1-8, the smoothed regression
line (for the 75th percentile HUC-level trend) has a positive slope, denoting that higher 75th
29 The LOESS procedure involves fitting of a least-squares curve based on application of a smoothing (or
bandwidth) parameter. Following the example provided by the SAB, we used a smoothing parameter of 0.2 in our
application of the LOESS procedure. The LOESS-based curve presented in Figure 1-8 reflects exclusion of two
outliers.
27
-------
percentile values are associated with higher HUC-level sample sizes.30 Interestingly, the mean
50th percentile HUC-level statistic across sample size strata (see Table 1-3) also tends to be
higher with more samples, but the trend is not as substantial or consistent as for the 75th
percentile estimates. This is expected because the median is a central tendency statistic, and
therefore any specific sample is likely to be closer to the central tendency than to other
percentiles. As a result, the small sample size issue is not anticipated to introduce significant
bias in the context of generating median estimates, but it may underestimate higher-end values
such as the 75th percentile. Uncertainty related to sample size and potential bias in the 75th
percentile HUC-level fish tissue Hg concentrations is also discussed in section 2.7, Table 2-15.
Interestingly, we do not see a consistent trend across the sampling site strata (i.e., the fish
tissue Hg statistic does not increase in a consistent manner with increasing number of sites - see
Table 1-2). This is not unexpected since the number of sampling sites is not directly correlated
with the number of observations at a given watershed (e.g., there are watersheds that have a few
sites each with a large number of observations and there are watersheds with a larger number of
sites, but each with only 1-2 observations).
HUC Frequency by Number of Hg Fish Tissue Samples
Frequency
(# of HUCs)
Illl
!••••
•V "V •>> t, <, fe -\
Number of Hg Fish Tissue Samples in HUC*
* When sample sizes a re 20 or greater, a category is used i.e. 20 corresponds to 20 to 25,25 corresponds to 26 to 30, etc.
Figure 1-7 Histogram Characterizing Frequency of Sample Sizes Across HUCs Included in
the Risk Assessment (illustrates fraction of HUCs with small sample size of 1-2)
30 This trend (higher sampled HUCs having higher 75th percentile fish tissue Hg values) could simply reflect the fact
that some states may target sampling at locations believed to have higher mercury impacts based on previous
sampling.
28
-------
E
Q.
Q.
OJ
=3 Cvj
c
OJ
(1)
Q.
T3
01
co O
OJ
-M
TO
0
bartdwidlh =
50 100 150
Number of observations per watershed
200
Figure 1-8 LOESS (locally-weighted scatter plot smoothing) Least-Square Regression of
-th
75 Percentile HUC-level Fish tissue Hg Levels Against HUC-level Sample Size
Table 1-3 Summary of 50th Percentile Hg Concentrations in Fish Tissue Samples by
Number of Observations and Number of Sites per HUC
Number of
Observations per HUC
1 Observation
2 Observations
3-5 Observations
6-10 Observations
> 10 Observations
Total
Number of sampling
sites per HUC
ISite
2 Sites
3-5 Sites
6-10 Sites
> 10 Sites
Total
Percentiles (of the 50th percentile HUC-level fish tissue
Hg concentration - ppm)
25th
0.077
0.08055
0.08403
0.1295
0.178
0.104
25th
0.1
0.1165
0.134761
0.100278
0.199375
0.104
50th
0.1564
0.1362
0.1443
0.207913
0.252444
0.19
50th
0.1815
0.2105
0.205034
0.175125
0.314303
0.19
75th
0.2887
0.25
0.235
0.36
0.376818
0.3203
75th
0.31
0.347335
0.32434
0.338674
0.41785
0.3203
90th
0.493
0.408
0.425
0.524
0.59
0.5033
90th
0.505
0.531
0.439024
0.491046
0.633312
0.5033
95th
0.67834
0.5235
0.639
0.72
0.728846
0.674
95th
0.6633
0.7371
0.496111
0.638407
0.633312
0.674
Mean
0.226803
0.195935
0.205501
0.274554
0.310015
0.25001
Mean
0.245554
0.269281
0.247729
0.236554
0.326412
0.25001
N
838
458
468
506
877
3147
N
2371
553
196
19
8
3147
There is also uncertainty due to excluding fish samples smaller than 7 inches.
Specifically, if subsistence fishers also target fish smaller than 7 inches in (and if those smaller
fish have lower MeHg concentrations), then the risk estimates could be high-biased for this
29
-------
group of subsistence fishers, since we excluded those smaller fish in our modeling. However, it
is important to reiterate that this analysis is not intended to generate a representative assessment
of the distribution of exposure and risk across the entire set of higher consuming self-caught
freshwater fishers. Given the goal of the analysis is to determine whether the potential exists for
adverse health impacts linked to self-caught fish consumption, it is reasonable (and indeed a
stated element of the scope of the analysis - see section 1.2) to focus on those behaviors by
subsistence fishers that would place them at greater exposure and risk. Given this goal, it is
reasonable to assume that a subset of subsistence fishers could focus on larger fish (i.e., > 7
inches) as they attempt to supplement their diet. By basing exposure and risk modeling on the
75th percentile fish and excluding fish smaller than 7 inches, we have targeted this subset of
subsistence fishers in this risk assessment.
While a reasonable assumption is that a subset of high-consuming subsistence fishers will
choose to catch and eat fish larger than 7 inches, reflecting recommendations by SAB, we
include a sensitivity analysis to examine the degree to which exclusion offish smaller than 7
inches affects risk estimates. Table 1-4 compares trends in different HUC-level percentile fish
tissue Hg concentrations (and the mean) based on consideration for (a) only fish > 7 inches in
length (used in the national-scale mercury risk assessment), and (b) fish of all lengths. Of
particular interest are the bolded rows in the table that provide results based on the 75th percentile
sample, which is the value used in the risk assessment. The results provided in Table 1-4 suggest
that excluding fish smaller 7 inches did not have a large impact on the HUC-level percentiles of
fish tissue Hg. This can be seen by comparing the 75th percentile HUC-level estimates using only
the fish larger than 7 inches with the 75th percentile estimates based on all fish combined. For
example, if we compare the average of the HUC-level 75th percentile values when only fish >7
inches in length were included (0.32 ppm) with the average of the HUC-level 75th percentile
values when all lengths are included (0.31 ppm) we see that the two values are similar. This
observation holds even if we look at higher end percentiles of the HUC-level 75th percentile
values (e.g., the 95th percentile of the HUC-level 75th percentile values are 0.89 ppm and 0.86
ppm, for the >7 inch and all lengths included, datasets, respectively). One likely reason that the
exclusion offish smaller than 7 inches did not have a large impact on the 75th percentile HUC-
level estimates is that this subset of smaller fish represents a relatively small fraction of the entire
fish tissue sample dataset considered for this analysis (i.e., -3,400 measurements for <7 inch fish
versus 35,600 measurements for > 7inch fish, or roughly ten times more samples for >7 inch
fish). Because the impact on fish tissue concentrations is small, we do not carry forward this
sensitivity analysis through the calculation of risks, as the impact on risk is also likely to be
small.
30
-------
Table 1-4 Comparison of HUC-Level Fish Tissue Hg Statistics for (a) Fish Tissue Dataset
with Fish >7 Inches and (b) Dataset with All fish Lengths Included
HUC-level statistic
N
Fish tissue Hg concentration (ppm) - as relevant
Mean
25th %
50th %
75th %
90th %
95th %
99th %
(a) Fish Length > 7 Inches (as used in the national-scale Hg risk assessment)
mean
25th %
50th %
75th %
90th %
Number of samples per HUC
Number of species per HUC
Number of sample sites per
HUC
3,147
3,147
3,147
3,147
3,147
3,147
3,147
3,147
0.27
0.19
0.25
0.32
0.40
11.23
2.69
1.41
0.11
0.08
0.10
0.13
0.14
1
1
1
0.20
0.14
0.19
0.25
0.29
4
2
1
0.35
0.25
0.32
0.42
0.53
12
4
1
0.54
0.40
0.50
0.67
0.86
28
6
2
0.71
0.52
0.67
0.89
1.13
44
7
o
6
1.09
0.89
1.06
1.36
1.86
106
10
5
(b) All fish lengths
mean
25th %
50th %
75th %
90th %
Number of samples per HUC
Number of species per HUC
Number of sampling sites per
HUC
3,268
3,268
3,268
3,268
3,268
3,268
3,268
3,268
0.25
0.18
0.24
0.31
0.39
11.83
2.85
1.41
0.10
0.07
0.10
0.12
0.13
1
1
1
0.19
0.13
0.18
0.23
0.27
4
2
1
0.33
0.24
0.30
0.40
0.51
13
4
1
0.52
0.38
0.48
0.65
0.83
29
6
2
0.68
0.50
0.65
0.86
1.10
46
8
3
1.05
0.85
1.02
1.34
1.82
123
11
5
1.4.2.1 Projecting 75th percentile fish tissue Hg concentrations for the 2016 scenario
Once 75th percentile fish tissue Hg concentrations are generated for each of the
watersheds, we then adjust these estimates to represent 75th percentile fish tissue Hg
concentrations for the 2016 scenario which are used in modeling risk for this scenario. As
mentioned in section 1.4 (Step 4 associated with Figure 1-2), projection offish tissue Hg
concentrations for 2016 is based on application of the proportionality assumption combined with
Hg deposition estimates for the 2005 and 2016 scenarios. Specifically, for a given watershed, we
multiply the 75th percentile measured fish tissue Hg concentration, by the ratio of total Hg
deposition in 2016 to total Hg deposition in 2005. This results in a projection of 2016 fish tissue
Hg concentrations. That 2016 projected total fish tissue Hg concentration is then used to model
HQ for the 2016 scenario in that watershed. Detail on the proportionality assumption (supported
by Mercury Maps simulations and more recent literature) is presented in section 1.4.6.
1.4.3 Defining subsistence fisher scenarios to include in the analysis
As described in section 1.4, defining subsistence fisher scenarios included in this analysis
requires three tasks: (a) identify subsistence fisher populations, (b) assess where they might be
11
-------
active and (c) define fish consumption rates for female subsistence fish consumers associated
with those populations. These three tasks are described below.
Identify subsistence fisher populations
As discussed in sections 1.2 and 1.4, this analysis estimates risk for scenarios
representing female subsistence fish consumers who have the potential to consume fish caught at
inland freshwater locations, since these populations are expected to experience the greatest U.S.
EGU-attributable risks. Therefore, in reviewing studies of fishing behavior, we prioritized
surveys characterizing fishing activity for populations with these characteristics. In addition, we
needed studies that provided statistically rigorous estimates of annual-average daily fish
consumption rates. Subsistence levels offish consumption are likely only experienced by
relatively small fractions of the study populations surveyed (e.g., 90th to 99th percentile
consumption rates).
Although we reviewed many studies, we ultimately identified three studies that met our
criteria for characterizing subsistence-level self-caught freshwater fish consumption. These three
studies included several subsistence populations differentiated by ethnicity and SES status
including: (a) White and Black populations (including female and low income strata) surveyed in
South Carolina (Burger et al., 2002), (b) Hispanic, Vietnamese and Laotian populations surveyed
in California (Shilling et al., 2010) and (c) Great Lakes Tribal populations (Chippewa and
Ojibwe) active around the Great Lakes (Bellinger et al., 2004).
These studies were used to characterize behavior for female subsistence fish consumers
assumed to be associated with the subsistence fisher populations described in the studies.31
Specifically, based on these studies, we characterized behavior for seven distinct female
subsistence fish consumer scenarios which were included in the risk assessment. These scenarios
are listed (and briefly described) in Table 1-5.
Table 1-5 Spatial Extent and Number of HUCs Reflected in Risk Modeling for the Female
subsistence fish consumer Scenarios Included in the Risk Assessment
Female
subsistence
fish consumer
Scenario
Typical
(non-SES
differentiated)
Low income
Description
Generalized (non-SES differentiated) female subsistence
fish consumers associated with fishing activity that
could occur at any of the watersheds across the nation
with fish tissue Hg data (i.e., source population
construct is not applied here).
Low income White female subsistence fish consumers
Spatial Extent
National (all HUCs
with fish tissue Hg
measurements)
Southeast
Number of
watersheds
(HUC12s)
Assessed For
Risk
3,141
860
31 With the exception of the Burger et al., 2002 study, which did provide consumption rates for females, we assumed
that rates provided for specific SES-differentiated populations could apply to pregnant females associated with those
fishers (e.g., either as family members or acquaintances). We recognize that there is uncertainty associated with this
assumption, but would also point out that the female consumption rates provided in Burger et al., 2002 (particularly
for the higher percentiles) are in the range of values seen for males and for non-sex differentiated survey groups both
in that study and in the other two studies of high-end self-caught fish consumption cited here.
32
-------
Female
subsistence
fish consumer
Scenario
White
(southeast)
Low income
Black
(southeast)
Low income
Hispanic
Great Lakes
Tribal
Laotian
Vietnamese
Description
active in the Southeast: in watersheds with fish tissue
Hg data where there are at least 25 White individuals
below the poverty line.
Low income Black female subsistence fish consumers
active in the Southeast in watersheds with fish tissue
Hg data where there are at least 25 Black individuals
below the poverty line.
Low income Hispanic female subsistence fish
consumers active at any watersheds in the continental
U.S. with fish tissue Hg data where there are at least 25
Hispanic individuals below the poverty line.
Tribal (Chippewa/Ojibwe) female subsistence fish
consumers active at any watersheds with fish tissue Hg
data within the ceded territories in the vicinity of the
Great Lakes (i.e., source population construct not
applied here).
Laotian female subsistence fish consumers active at any
watersheds in the continental U.S. with fish tissue Hg
data where there are at least 25 Laotians.
Vietnamese female subsistence fish consumers active at
any watersheds in the continental U.S. with fish tissue
Hg data where there are at least 25 Vietnamese.
Spatial Extent
Southeast
National (where
there are low
income Hispanics)
Vicinity of Great
Lakes (ceded
territories)
National (where
there are Laotians)
National (where
there are
Vietnamese)
Number of
watersheds
(HUC12s)
Assessed For
Risk
756
1,325
361
131
336
Assess where the subsistence fisher populations might be active
The following approach was used to identify the spatial extent of activity for each of the
seven female subsistence fish consumer scenarios. For all of the scenarios (with the exception of
the Tribal scenario associated with Great Lakes), we assumed that high-end fishing behavior
could be generalized beyond the specific geographic areas covered in a particular study. This
type of generalization was necessary to provide sufficient coverage for the continental U.S. and
in particular, the eastern part of the U.S. where U.S. EGU-attributable deposition is higher and
where we have more measured Hg fish tissue data. In deciding how to extend coverage for each
fisher population, we considered several factors including (a) the potential for high-end fishing
activity to be culturally-related and therefore more likely to be followed by populations of a
given ethnicity living across the U.S. and (b) the potential that subsistence fishing activity (and
related consumption) might be driven by economic need. Because we assumed that fishing
activity is culturally related, we generalized fishing activity by Hispanics, Laotians and
Vietnamese beyond California, where the study providing consumption rates was conducted, to
the nation. Similarly, we assumed that surveys of high-end fish consumption by low income
Whites and low income Blacks in South Carolina might be generalized to the southeast,
reflecting similar cultural practices by these groups within that region.
Once we identified the regions of the country where a particular subsistence fisher
population might be active, we then identified the specific subset of watersheds within that
region where the high-end fishing activity (and consumption by female subsistence fish
consumers associated with those subsistence fishers) might occur. This task was challenging
33
-------
because we do not have data characterizing the number and location of subsistence fishers for
any of the fishing populations under consideration. Therefore, we developed and applied a
source population construct to guide identification of areas (watersheds) where a given
subsistence scenario might be active.32 This approach requires that in order for a subsistence
fisher scenario to have the potential to be active at a given watershed, that watershed must be in
close proximity to a source population matching the ethnic/SES composition of the subsistence
fishers being represented. In this case, we specified that a watershed would have to intersect a
U.S. Census tract with at least 25 individuals with the same ethnic/SES composition as the
subsistence fisher scenario (i.e., a source population) for there to be the potential for that type of
fishing at that watershed. So, for example the Laotian subsistence fisher scenario was only
assessed at those watersheds intersecting U.S. Census tracts with at least 25 Laotians. In the case
of low income White and Black subsistence fishers in the southeast as well as Hispanics assessed
nationally, we required that the watershed intersect U.S. Census tracts with at least 25
individuals from that ethnic group who fall below the poverty line. The SAB peer review panel
agreed that the criterion of using 25 persons within a source population to identify watersheds
with the potential for subsistence fishing activity is a reasonable approach. The specific approach
used to extrapolate coverage regionally and then determine the subset of watersheds to include in
risk modeling for each subsistence fisher scenario is described both in Table 1-5 and in greater
detail in Table 1-6 (in Table 1-6 see columns "C" and "E"). As described in Table 1-6 (columns
"C" and "E"), the approach used for the Tribal population and typical female subsistence fish
consumer scenarios differs from those used for the other scenarios. In the case of the Tribal
subsistence fisher scenario, given the potential for fishing activity to be closely associated with
heritage cultural practices, we have not extrapolated fishing activity for the Chippewa/Ojibwe
outside of their ceded territories, although we do assume that all watersheds within those
territories with fish tissue Hg data have the potential for fishing activity by this group.
The typical female subsistence fish consumer scenario differs from the other scenarios in
that it is not SES-differentiated and is assessed nationally at all watersheds with fish tissue Hg
data (i.e., the source population construct is not used to constrain its spatial extent). The
generalized nature of this scenario means that it provides broader coverage for subsistence fisher
activity and for this reason we assume that this scenario (together with female subsistence
consumption associated with this fishing activity) could potentially occur at any watershed with
fish tissue Hg data.33 As noted in section 1.2 and 1.4, because the typical female subsistence fish
consumer is the most generalized of the scenarios and has the greatest spatial extent, we place
greater emphasis on this scenario in presenting risk estimates.34 Application of the typical female
32 As described in section 1.4, a source population is a group of individuals with demographic attributes matching
those of the subsistence fisher group being evaluated. While we cannot enumerate the subsistence fishers directly,
we can use demographic data to determine if the underlying source population is present in the vicinity of a
watershed with fish tissue Hg data.
33 We are not suggesting that we expect the scenario to occur as a certainty at all watersheds, or that we can predict
the level of activity (number of fishers or consumers) at each watershed. Rather, for this analysis, we are stating that
we believe it reasonable to assume that the typical female subsistence fish consumer scenario (and associated fishing
activity) could potentially occur at some subset of the watersheds with fish tissue Hg data.
34 The typical female subsistence scenario included in this revised risk assessment is similar to the high consuming
female scenario included in the March version of the Mercury Risk TSD. However, the typical female subsistence
scenario is applied without consideration for a source population, while in the case of the high-consuming female
angler; we did consider a source population based on poverty. Because the typical female subsistence scenario does
not use a source population (and is applied uniformly to all watersheds with fish tissue Hg data except for those
34
-------
subsistence fish consumers scenario to all watersheds with fish tissue Hg data also helps to
address the concern raised by the SAB that the March version of the national-scale Hg risk
assessment may have excluded more remote water bodies that could be fished by subsistence
anglers (leading to an underestimation of the percent of watersheds where Hg exposure from
U.S. EGU sources is a risk). As mentioned earlier, Table 1-5 lists the seven female subsistence
fish consumer scenarios and presents the number of watersheds assessed for risk for each
scenario.
Define self-caught fish consumption rates for the subsistence scenarios
After identifying the set of female subsistence fish consumer scenarios to include in the
analysis and determining where those scenarios should be applied, next we defined high-end
(subsistence) self caught fish consumption rates for those scenarios. The three studies referenced
in Table 1-6 provided either subsistence-level consumption rates at the 90th to 99th percentile, or
the statistical parameters necessary to calculate those percentiles (e.g., median and standard
deviations).35 In establishing consumption rates for the seven female subsistence fish consumer
scenarios, we either used rates from the studies directly (as in the case of the typical female
subsistence fish consumer which uses female consumptions rates from the Burger et al., 2002
study) or we used non-sex differentiated consumption rates, assuming that they applied to
women of childbearing age. As noted earlier in this section, there is uncertainty associated with
the assumption that non-sex differentiated consumption rates could apply to women of child-
bearing age (see footnote 33).
The subsistence-level consumption rates used in modeling risk for these scenarios are
presented in Table 1-6, along with notes relevant to their interpretation in the context of this risk
assessment. Of particular interest was whether the consumption rates represented annual-
averages of daily "as consumed/prepared" values. As noted earlier in 1.4, this issue is critical to
ensuring the appropriate technical linkages between the types offish tissue Hg measurements
made, application of preparation adjustment factors and ultimately, calculation of exposure and
risk. This issue is discussed in greater length in the next section.
The fish consumption rates presented in Table 1-6 clearly represent subsistence behavior
since they reflect an individual making significant contributions to their diet through self-caught
fish consumption. In most instances, the rates are equivalent to between one 8oz fish meal every
few days to a larger fish meal (12oz or more) every day. While these consumption scenarios are
identified as "subsistence" due to their magnitude, they could be experienced by individuals
purposefully supplementing their diet due to economic need or due to high levels of recreational
fishing activity. While these scenarios could represent either subsistence or high-end recreational
fishing, the resulting exposure and risk characterization would be similar. For typical female
subsistence fish consumers, as mentioned earlier, this scenario is applied uniformly across all
watersheds with fish tissue Hg data, and whether we define consumption as subsistence or high-
end recreational fishing does not affect the exposure and risk characterization. In either case, the
excluded due to potential impacts from non-air deposition), this scenario does provide greater coverage
geographically than did the high-consuming female which was only applied to watersheds with fish tissue Hg data
and at least 25 members of the source population (individuals living below the poverty line).
35 In those instances where a specific percentile was not provided, we estimated that value using the statistical
parameters provided together with the assumption that the underlying fish consumption distribution was lognormal.
35
-------
risk estimates represent risk experienced by an individual residing in the vicinity of a particular
watershed who consumes self-caught fish at a subsistence level.36 In addition to the three studies
we used to define subsistence fish consumption rates, we also reviewed additional studies
characterizing higher-levels of self-caught fish consumption in the U.S. While these additional
studies had limitations that prevented their use in this risk assessment, they generally support the
rates of self-caught fish consumption modeled in the analysis.
A subset of these "supporting" studies is briefly described below:
• A study by Burger et al., 1999, examined recreational and subsistence fishing activity
along the Savannah river in Georgia to determine the role played by socio-economic
status (SES) factors (including race, education and income) and determined levels of self-
caught fish consumption in this study area. The study suggested that all three factors are
associated with levels of fishing activity. Specifically, in the case of race, the study
showed that Blacks tend to have much higher rates offish consumption than Whites. For
both groups, the study suggested that upper-end percentile consumption rates could be
high enough to approach subsistence levels. For example, a -200 g/day fish consumption
rate represented the 98th percentile for Whites, but only the 92nd percentile for Blacks.
This study supports the presence of high-end consumption rates for both Blacks and
Whites that approach or meet subsistence levels in this area of the country. While this
study did provide support for high self-caught fish consumption rates and SES
differentiation offish consuming populations, because of its geographically limited scope
(i.e., focus on Savannah River), we elected not to use it directly in defining consumption
rates for the risk assessment.
• A study by Moya et al., 2008 examined factors associated with regional differences in
patterns offish consumption, including age, ethnicity (including Tribal affiliation),
socioeconomic status (e.g., income, education), and type/source offish consumed
(freshwater, marine, and estuarine obtained from commercial sources versus self-caught).
The study examined fishing activity in four states (i.e., CT, FL, MN, and ND) and
provided estimates of high-end self-caught fish consumption for populations. Higher,
subsistence-level consumption rates were identified for fishing populations in FL, MN
(specifically for Tribes) and CT (for Asian populations, although it is not clear whether
the rates for Asians are for self-caught fish consumption). Higher-end rates reported for
ND and for general fishers in CT and MN did not approach the range of subsistence
levels of consumption. However, the study designs used in these surveys may not
effectively capture the small fraction of the overall population likely engaging in high-
end subsistence levels of self-caught fishing behavior. This study supports the existence
of subsistence fishing populations in FL and for Tribes in MN. However, failure to
capture similar behavior in ND and CT does not necessarily suggest that this type of
36 To generate population-weighted risk estimates, it is important to differentiate between high-end recreational and
necessity-based subsistence activity, since the associated populations at any given watershed could differ, leading to
different population-weighted risk distributions. However, because we lacked other data needed to calculate
population-weighted risks, this differentiation is not relevant.
36
-------
behavior is non-existent, but it may suggest that subsistence behavior is less prevalent
than in FL. We did not elect to use the data for FL to define consumption rates since we
already had the Burger et al., 2002 study which covers a region of the Southeast (South
Carolina) where we have a larger number of watersheds being modeled for risk.
37
-------
Table 1-6 Fish consumption rates and additional behavior-related information for subsistence populations included in the
analysis
(A) Fish consuming
populations covered by
study (and reference
information)
(B) Overview of
study
(C) Assumption regarding where
the subsistence fisher scenario
(and associated female subsistence
fish consumers) might be active
(D) Self-caught fish consumption rates
(mean, 90th, 95th, 99th) g/day and notes on
type of consumption rate
(e.g., temporal averaging period and as
purchased versus as consumed)
(E) Notes on consumption rate
information relevant to the risk
assessment
Higher self-caught fish
consuming populations
(White, Black and female)
surveyed in South
Carolina
Citation: Daily
consumption of wild fish
and game: Exposures of
high end recreationalists,
Burger et al., International
Journal of Environmental
Health Research, 12:4, p.
343-354, July, 2002
Random survey of
participants in the
Palmetto
Sportsmen's Classic
in Columbia SC
(1998). Population
interested in
fishing/hunting (not
general population -
represents outdoor
enthusiasts in SC)
- the Black and White fisher
populations were extrapolated to
cover watersheds modeled for risk
in the Southeastern states. The
rationale for this was that fishing
activity by these two groups could
be generalized in this region of the
country. These scenarios were only
assessed for watersheds in the
Southeast located within U.S.
Census tracts with at least 25
individuals from that ethnic group
below the poverty line.
- given the focus of the risk
assessment on consumption by
women (in considering exposures
for pregnant women in particular),
we extrapolated the typical female
consumer scenario to all watersheds
in the continental U.S. and, given
the more generalized nature of this
scenario (no ethnic or SES
differentiation), we assessed the
scenario for all watersheds included
in the risk assessment (i.e., we did
not apply the source population
criterion used for the other
scenarios).
-Black: 171, 446, 557, NC *
-White: 38.8,93,129,286
-female: 39.1,123,173,373
* the sample size for this population is only
39, reducing overall confidence in a 99th
consumption rate (therefore, this high-end
percentile was not included in the risk
assessment)
Consumption rates are annual-average values
expressed as meals or portions (as prepared)
(median, 75th, 90th, 95th, and 99th). Survey
asked respondents by month for number of
meals of different type of fish and serving size
(here models were used to demonstrate
different meal or serving sizes). Authors
could then use this to estimate monthly
consumption rates and convert these into an
annual average.
Sample size is variable - out of 458
respondents, 39 are Blacks, 149 are
female and 98 are low income -
Black n is relatively smaller than the
other groups, which increases
uncertainty in higher percentile
values provided for this group.
The authors point out that these
results highlight the considerable
spread between high-end consumers
and more typical behavior (95th
percentile is more than 10X greater
than the mean or median intake rate
for wild-caught fish).
Results are also provided for low
income (0-20KS annual income).
These consumption rates are
relatively high particularly for the
higher percentiles (90th, 95th and 99th
rates are: 285, 429 and 590 g/day).
This observation forms the basis for
our decision to assess a number of
the subsistence populations only for
watersheds located in U.S. Census
tracts containing members of source
populations below the poverty line
for the White and Black populations.
38
-------
(A) Fish consuming
populations covered by
study (and reference
information)
(B) Overview of
study
(C) Assumption regarding where
the subsistence fisher scenario
(and associated female subsistence
fish consumers) might be active
(D) Self-caught fish consumption rates
(mean, 90th, 95th, 99th) g/day and notes on
type of consumption rate
(e.g., temporal averaging period and as
purchased versus as consumed)
(E) Notes on consumption rate
information relevant to the risk
assessment
Higher self-caught fish
consuming ethnic
populations including
Hispanics, Laotians and
Vietnamese surveyed in
California
Citation: Contaminated
fish consumption in
California's Central Valley
Delta (Shilling et al.,
Environmental Research
110, p. 334-344(2010)
Study looks at
subsistence fishing
activity among
ethnic groups
associated with
more urbanized
areas near the
Sacramento and San
Joaquin rivers in the
Central Valley in
CA.
- the Hispanic fishing scenario was
extrapolated to cover watersheds
located in U.S. Census tracts with at
least 25 low income members of the
ethnic populations (e.g., the
Hispanic consumption rates would
be applied to the subset of the 3,141
watersheds located in U.S. Census
tracts with at least 25 low income
Hispanic individuals).
- the Laotian and Vietnamese
fishing scenarios were extrapolated
to cover watersheds located in U.S.
Census tracts with at least 25
members of the underlying ethnic
group.
- Hispanic: 25.8, 98, 155.9, NC*
-Lao: 47.2, 144.8, 265.8, NC*
- Vietnamese: 27.1, 99.1, 152.4,NC*
* 95th percentile values were provided in the
study. 90th percentile values were calculated
using Crystal Ball (based on the median and
standard deviations provided) assuming a log-
normality of the consumption rate
distributions. 99th percentile consumption rates
were not provided (or derived) for any of these
populations due to small sample sizes of the
study populations.
Consumption rates are annual-average
estimates (cooked weight - as filets) based on
the fact that they compare their average values
against other rates from the literature
including rates used by the EPA in the
regulatory context - all of which are annual
averages. Survey used different sized models
of cooked fish filets.
The authors note that many of these
ethnic groups relied on fishing in
origin countries and bring that
practice here (e.g., Cambodian,
Vietnamese and Mexican). The
authors also note that fish
consumption rates reported here for
specific ethnic groups (specifically
Southeast Asian) are generally in-
line with rates seen in WA and OR
studies.
39
-------
(A) Fish consuming
populations covered by
study (and reference
information)
(B) Overview of
study
(C) Assumption regarding where
the subsistence fisher scenario
(and associated female subsistence
fish consumers) might be active
(D) Self-caught fish consumption rates
(mean, 90th, 95th, 99th) g/day and notes on
type of consumption rate
(e.g., temporal averaging period and as
purchased versus as consumed)
(E) Notes on consumption rate
information relevant to the risk
assessment
High-end self-caught fish
consuming Chippewa and
Ojibwa Tribal populations
active in the vicinity of the
Great Lakes.
Citation: Exposure
assessment and initial
intervention regarding fish
consumption of tribal
members in the Upper
Great Lakes Region in the
United States. Bellinger,
Environmental Research
95 (2004) p. 325-340
This study
contrasted self-
reported fish
consumption rates
by Tribes in the
Great Lakes area
with "actual" fish
consumption rates
collected for a
subset of the
original study
population (147 of
822 from 4 Tribal
population/location
combinations). The
study found that
actual fish
consumption rates
were lower than
reported values.
Activity only assumed to occur in
areas ceded to the Tribes covered in
the study (regions in the vicinity of
the Great Lakes). Because fishing
activity is highly variable across
Tribes (and closely associated with
heritage cultural practices) we have
not extrapolated fishing behavior for
these Tribes outside of the specific
populations and regions covered.
- reported value for all Tribal areas (in the
study) combined: 62, 136.2,213.1,492.8
All higher percentiles (90th - 99th) were
derived using Crystal Ball (based on median
and standard deviations and an assumption of
log-normally distributed variability in
consumption rates)
Consumption rates appear to be annual
average values. Study includes reference to
querying for the number of fish meals in a
year, which suggests that estimates are annual-
averages (but it is not explicitly stated).
Ingestion rates are in-line with other high-end
consumer rates cited in this table. It also
appears that estimates are for amount
consumed (i.e., meal weight), based on the
fact that the "actual" estimates provided in the
article focused on this type of consumption
rate.
40
While the "actual" consumption
rates collected for a subset of the
families were far lower than the
reported values (often an order of
magnitude smaller), a number of
factors resulted in a decision to use
the reported values rather than the
actual values in the risk assessment.
First, and most importantly, the
sample size is very small for the
"actual" analysis with n's ranging
from 12 to 54 individuals
(representing a smaller number of
associated families) for the different
survey groups. These small
sampling rates reduce the
probability of capturing individuals
with higher consumption rates in the
broader population. It also appears
that the actual values may cover
walleye specifically and not include
all fish, which could bias these
values downward. There is concern
that, even if consumption rates have
decreased, actual heritage cultural
practices could still exist (or there
could be a desire to return to those
rates), in which case, risks levels
associated with those higher
historical consumption rates could
be important to assess. And finally,
the high-end percentile consumption
rates derived based on reported
mean consumption rates (and
standard deviations) are in-line with
subsistence consumption rates seen
for other populations in the U.S.
Therefore, these Tribal high-end fish
consumption rates would general
comport with subsistence fish
consumption activity and therefore
are considered reasonable to include
in the risk assessment.
-------
1.4.4 Estimating total fish consumption-related MeHg exposure (2016 scenario)
Next, we estimated total exposure to Hg for the set of seven female subsistence fish
consumer scenarios being evaluated (for the 2016 scenario). To generate estimates of total Hg
exposure, we combined the 75th percentile fish tissue Hg values (projected for 2016) with three
exposure-related factors including: (a) the consumption rates for a subsistence fisher scenarios in
grams offish per day, (b) a conversion factor between total Hg measurements in fish and MeHg
levels in fish and (c) a food preparation adjustment (cooking fish can increase MeHg
concentrations).
In modeling MeHg exposure through fish consumption, it is important to verify that the
linkages between factors used in modeling exposure (i.e., the variables in the exposure equation
listed above) are conceptually correct. For example, we need to convert the total Hg levels in fish
to MeHg levels in order to calculate the HQ, which is based on MeHg intake. Similarly, if fish
ingestion rates are specified in terms of "as cooked" or "as consumed" rather than "as
purchased" or "as market basket," then we must factor in cooking/preparation in specifying the
fish tissue Hg concentrations. Failure to correctly link these exposure factors can result in biased
estimates of MeHg intake and consequently biased risk estimates. Below we discuss each
exposure factor, including the data used to specify that factor and the linkage of that factor to
others in the exposure equation. Then, we present the exposure equation.
Mercury fish tissue concentration (FTC): We calculated the 75th percentile fish tissue Hg
estimate for each watershed and then project those to represent the 2016 scenario (as described in
section 1.4.2.1). Generally, measured fish tissue Hg is in terms of total Hg. Because we need
MeHg levels in fish, we convert total Hg levels measured in fish into MeHg levels using the
mercury conversion factor (MCF).
Mercury conversion factor (MCF): We applied a factor of 0.95 (unitless) to convert total Hg
levels in fish tissue samples into MeHg levels. This conversion factor is based on conclusions
reached in the Mercury Study Report to Congress (MRTC) (U.S. EPA, 1997) that more than
90% of Hg in fish is MeHg. The conclusion presented in the MRTC is in turn based on two
studies (Bloom, 1992 and Morgan et al., 1994). The 0.95 factor represents a median value
between an assumption that all of the Hg in fish is MeHg and the lower bound value of 0.90 cited
in the MRTC document.
Food preparation/cooking adjustment factor (FPCAF): Because we use fish consumption
rates for "as consumed" portions, we need to adjust the fish tissue Hg concentrations to reflect
concentrations after food preparation. Cooking fish typically increases MeHg levels per unit fish
because Hg concentrates in the muscle while preparation involves removal primarily of non-
muscle elements of the fish (e.g., water, fat etc) (Morgan et al., 1997). The FPCAF factor used in
the analysis is 1.5 (i.e., an increase of 50% in the MeHg concentration per unit fish due to
preparation/cooking). This factor is based on the Morgan et al., 1997 study, which estimated
factors of between 1.1 and 1.5 for walleye and 1.5 and 2.0 for lake trout.
The SAB recommended that additional studies be acknowledged as providing information
regarding the FPCAF, citing two alternative studies (Farias et al., 2010 and Musaiger et al.
2008). After assessing these studies, we conclude that they do not support a lower
41
-------
preparation/cooking loss adjustment factor. The Farias study appears to suggest that preparing
fish by Manaus residents in the Amazon could decrease Hg concentrations. Although this
initially suggests that the 1.5 factor used in the risk assessment is biased high, a closer read of the
Farias study suggests that the authors may have measured non-fish components added to dishes
(e.g., onions, heavy breading etc) in post-cooking measurements, which could provide the
appearance of a cooking loss in Hg while actual fish tissue Hg concentrations could have
increased. The Musaiger et al. 2008 study compared Hg levels across different types offish
meals after preparation rather than pre and post-cooking Hg levels. In fact, the authors state that
cooking is not a means of reducing Hg because it typically removes fat and water, while Hg is
located in the meat and therefore will not be reduced.37
Fish consumption rate (FCR): We reviewed the three studies providing fish consumption rates
used in the risk assessment to verify that consumption rates (a) are annual-averages and not rates
reflecting elevated consumption during shorter seasonal periods and (b) represent "as consumed"
rather than "as purchased" values. Based on the review of the three studies we concluded that the
consumption rates are annual average daily consumption rates for "as cooked/as prepared." In
some instances, a study did not explicitly state whether rates were annual-averages and/or "as
cooked", and we had to infer this from other information provided in the article. Details
regarding our assessment of each of the three studies is provided in Table 1-6 (column D).
The following equation shows how these factors were combined to generate estimates of
annual-average daily MeHg exposure per kg body weight:
IR = FTC(2016) * MCF * FPCAF * FCR
BW
where,
IR = daily MeHg intake rate (ug/kg-day)
FTC(2016) = Hg fish tissue concentration (ug/g or ppm of total Hg) projected for the
2016 scenario
MCF = Hg conversion factor (unitless)
FPCAF = food preparation/cooking adjustment factor (unitless)
FCR = fish consumption rate (g/day)
BW = body weight (kg)
37 As part of revising the national-scale Hg risk assessment, we completed a literature review focusing on the issue
of cooking/preparation of fish and the effect on Hg concentration. This literature review included the two studies
recommended by the SAB. That review did not identify any studies that argued against use of an adjustment factor
of 1.5 and consequently, we continue to use that factor in modeling exposure and risk.
42
-------
1.4.5 Estimating risk (RfD-based hazard quotient) (2016 scenario)
Once we estimate the MeHg intake rate for each female subsistence fish consumer
scenarios at each watershed (for 2016), we compare these exposure estimates to the MeHg RfD
to generate HQ estimates. The MeHg RfD is 0.0001 mg/kg-day (equivalent to 0.1 jig/kg-day)
and was published by the EPA in the Integrated Risk Information System in 2001 (US EPA,
200la) (http://www.epa.gov/iris/subst/0073.htm).
Reflecting precedent in interpreting HQ estimates, we consider exposures above the RfD
(i.e., an HQ above one) to represent a potential public health hazard. Consequently, watersheds
with female subsistence fish consumer exposures above the RfD are considered to have the
potential for consumers of self-caught fish from that watershed to experience a public health
hazard due to MeHg exposure. An important factor to consider in interpreting HQ's is the
precision underlying these risk estimates. While precision associated with the exposure estimates
may be higher, the RfD is only reported to one significant digit, which limits the precision in the
HQ estimate. Consequently, we interpret an HQ of 1.5 or greater as representing an exposure that
exceeds the RfD (since this value will round to two expressed as a whole number). Conversely,
exposures of 1.49999 or less are considered not to exceed the RfD given this rounding
convention for HQs.38
In response to SAB recommendations that the IQ loss endpoint may not fully capture the
range of neurodevelopmental effects associated with Hg exposure, we have deemphasized this
category of risk metrics and moved the IQ discussion to the appendices. We discuss the
approach used to generate IQ loss estimates including the concentration-response function in
Appendix A and we provide the results and approach used to interpret the public health
significant of IQ loss in Appendix B.
1.4.6 Estimation of U.S. EGU-attributable risk (2016 scenario)
Next, we estimated the fraction of total risk (HQ) that is attributable to U.S. EGUs (for
the 2016 scenario). The estimate of U.S. EGU-attributable risk is at the core of the 2-stage risk
characterization framework (see Section 1.3). We estimate the U.S. EGU-attributable fraction of
total risk using the linear proportionality assumption linking changes in Hg deposition over
watersheds with changes in fish tissue Hg concentrations.39 As noted in section 1.4.2.1, this
proportionality assumption states that, under steady state conditions, a change in Hg deposition
over a given watershed will result in a proportional change in fish tissue Hg concentrations and
associated exposure and risk. The proportionality assumption is supported by the Mercury Maps
38 Note, that for the U.S. EGU incremental contribution analysis, anHQ of less than 1.5 does not necessarily
indicate there is no public health hazard related to U.S. EGU emissions. Rather, it suggests that, for those specific
watersheds, we need to also consider whether total risk (i.e., the HQ reflecting total MeHg exposure) exceeds 1.5
and therefore represents a potential public health hazard. If that is the case, then we would consider the degree to
which U.S. EGUs contribute to that total exposure because incremental exposures above the RfD increase the risk.
39 Because risk at the HUC-level is linearly related to the fish tissue Hg concentration (see sections 1.4.4 and 1.4.5),
this proportionality assumption can be used to relate changes in mercury deposition to change in total risk at that
HUC.
43
-------
analysis (U.S. EPA, 2001b). In addition, a number of additional studies (some of which were
highlighted by the SAB in their review of the March TSD) support a proportional relationship
between changes in Hg deposition and changes in fish tissue Hg concentrations, although with
important caveats related to the temporal response of aquatic systems to changes in Hg
deposition/loading. There are also important criteria that must be met for the proportionality
assumption to hold, which we discuss below.
We used the CMAQ air quality model to estimate Hg deposition over U.S. watersheds for
all sources of mercury, including U.S. and non-U.S. sources. We also modeled Hg deposition
directly from U.S. EGUs by zeroing out Hg emissions from U.S. EGU sources and subtracting
the results from total Hg deposition to isolate the U.S. EGU contribution. The CMAQ modeling
for this analysis is described in section 1.4.6.3.
Hg deposition estimates at each watershed coupled with the proportionality assumption
described above can be used to estimate U.S. EGU-attributable HQ risk for the 2016 scenario. To
estimate the U.S. EGU-attributable fraction of that 2016 HQ estimate, we use the ratio of U.S.
EGU Hg deposition versus the estimate for total Hg deposition (both for the 2016 scenario in that
watershed). The step-wise procedure for completing the estimate of U.S. EGU-attributable risk
for 2016 is outlined below:
Generate U.S. EGU-attributable HQ risk estimates for the 2016 scenario: We use
the 2016 total and U.S. EGU-attributable CMAQ deposition results with the estimated
total HQ risk estimates for 2016 to estimate the U.S. EGU-attributable HQ risk at each
watershed as follows:
2016 EGU-HQ = 2016 THQ * (2016 EGU Hg dep / 2016 total Hg dep)
Where:
2016 EGU-HQ: U.S. EGU-attributable HQ risk for the 2016 scenario
2016 THQ: HUC-level total HQ risk for 2016 generated as described in section 1.4.5.
2016 EGU Hg dep: 2016 CMAQ-based projections of U.S. EGU-related Hg deposition
over a given watershed)
2016 total EGU dep: 2016 CMAQ-based projections of total Hg deposition over a given
watershed
1.4.6.1 Mercury Maps analysis
Results of the EPA's Office of Water's Mercury Maps analysis support the
proportionality assumption (U.S. EPA, 2001b). The Mercury Maps analysis used a simplified
form of the IEM-2M model applied in EPA's Mercury Study Report to Congress (U.S. EPA,
1997). By simplifying the assumptions inherent in the freshwater ecosystem models that were
described in the Report to Congress, the Mercury Maps model showed that these models
converge at a steady-state solution for MeHg concentrations in fish that are proportional to
changes in Hg inputs from atmospheric deposition. This steady-state solution only applies in
situations where air deposition is the only significant source of Hg to a water body, and the
physical, chemical, and biological characteristics of the ecosystem remain constant overtime.
44
-------
Consequently, the proportionality assumption used to estimate the U.S. EGU-attributable
fraction of risk would ideally only be applied to watersheds where these criteria have been met.
Application of the proportionality assumption in situations where these criteria have not been
met introduces uncertainty in the apportionment of total risk. EPA recognizes that concentrations
of MeHg in fish across all ecosystems may not reach steady state and that ecosystem conditions
affecting Hg dynamics are unlikely to remain constant over time. EPA further recognizes that
many waterbodies, particularly in areas of historic gold and Hg mining in western states, contain
significant non-air sources of Hg.40 Finally, EPA recognizes that Mercury Maps does not
estimate the time lag between a reduction in Hg deposition and a reduction in the MeHg
concentrations in fish.
In their peer review, the SAB noted that there are other modeling tools available to link
deposition to fish tissue concentrations, but did not consider them to be superior for this analysis,
nor did they recommend their use. The SAB specifically noted that the Regional Mercury
Cycling Model (R-MCM) could also be used for a national assessment, but they also noted that
the R-MCM is more data intensive and the results produced by the two model approaches should
be equivalent. We did not have all of the data inputs (water chemistry, methylation potential,
etc) that would be required to run the R-MCM, and given the SAB recommendation, and their
comment that "it is unlikely that substantial additional insight would be gained with the
alternative model framework," we elected to use the proportionality assumption as supported by
the Mercury Maps modeling and the peer-reviewed literature. Because the Mercury Maps
approach only applies in those watersheds where aerial deposition is the dominant source of Hg
loading, we excluded watersheds with substantial non-U.S. EGU anthropogenic emissions of Hg
from the risk assessment as described in section 1.4.2.
There are a number of limitations and uncertainties associated with the application of the
Mercury Maps approach in the context of this risk assessment. These limitations are discussed
here and addressed in Table 2-15. Applying Mercury Maps to apportion fish tissue Hg
concentrations and consequently exposure and risk between U.S. EGUs and all other sources of
Hg at the watershed-level assumes that the relationship between fish tissue levels and Hg
deposition has remained fairly consistent such that near steady-state conditions have been
reached. However, in reality, patterns of Hg deposition for the period during which the fish tissue
samples were collected (2000 to 2009) have not remained constant. In addition, those fish tissue
concentrations may actually reflect patterns of Hg deposition from earlier time periods (e.g., the
1990s) when Hg emissions from U.S. sources were experiencing substantial decreases.41 In
40 As described below, we have excluded those watersheds containing gold mines or with other non-EGU related
anthropogenic Hg emissions exceeding specified thresholds.
41 The discussion of model uncertainty provided in the Technical Support Document describing the Mercury Maps
analysis (MMAPs TSD, U.S. EPA, 2001) also addresses the fact that proportionality between a decrease in Hg
deposition and changes in fish tissue Hg concentrations may not be fully realized for some time due to the lagging
effect of Hg that has built up in sediment. Simulations discussed in the MMAPs TSD suggested that substantial
concentrations of Hg could build up in sediment with these loadings effectively buffering the impact of reductions in
Hg. This issue does not invalidate the application of the proportionality assumption in the national-scale mercury
risk assessment, but it does suggest that the full effect of predicted reductions in risk may not be seen in the near-
term (this issue is discussed further in the context of other studies in section 1.4.6.2). The discussion of model
uncertainty in the MMAPs TSD also addresses potential non-linearity in methylation. Citing the DiPasquale, et al,
2000 study, the MMAPs TSD notes the potential for reduced methylation at very high Hg loadings typically
experienced in mining areas. However, since our analysis excluded areas likely impacted by active gold mines, we
believe that this issue has been largely ameliorated in the context of the national-scale mercury risk assessment.
45
-------
addition, other factors that can affect rates of Hg methylation (e.g., sulfur deposition to
waterbodies, pH of the waterbodies) also have not remained constant over the past 1-2 decades
for most watersheds. The fact that many of these factors related to methylation in fish have not
remained constant introduces uncertainty into the application of the Mercury Maps based
proportionality assumption. However, we believe that the Mercury Maps approach for
apportioning fish tissue Hg concentrations is still appropriate to use, particularly if we are not
attempting to characterize the temporal response and instead, can assume that sufficient time has
passed for near steady state conditions to be reached (see section 1.4.6.2 below). Furthermore,
while we have excluded watersheds with substantial non-air loading of Hg from industrial
activity and mines, we did not consider municipal sewage emissions of Hg to watersheds. This
introduces additional uncertainty into the analysis, since watersheds could have a substantial
fraction of Hg loading originating from municipal sewage treatment. In these instances, failure to
consider this source of non-air loading could result in a high-bias in the estimates of the U.S.
EGU-attributable fraction of deposition (and hence risk), since the contribution from this non-air
source would not have been considered.
1.4.6.2 Additional research supporting the proportionality assumption and
examining the issue of temporal response
The SAB commented that several recent publications have supported the finding of a
linear relationship between Hg loading and accumulation in aquatic biota (Orihel et al., 2007;
Orihel et al., 2008; Harris et al., 2007). The SAB noted that these studies suggest that that Hg
deposited directly to aquatic ecosystems can become quickly available to biota and accumulated
in fish, and reductions in atmospheric Hg deposition should lead to decreases in MeHg
concentrations in biota.
EPA has reviewed the studies identified by the SAB, together with a study by Knightes et
al., (2009). All of these studies, to varying degrees, suggest that when we are considering
reductions in fish tissue Hg concentrations following reductions in Hg deposition to a watershed,
we are likely to see a 2-stage response including: (a) an initial more rapid phase of reduction
(ranging from months to a few years) reflecting a decrease in direct loading to the waterbody and
subsequent reductions in fish tissue Hg concentrations related to decreased water column Hg
levels and (b) a second slower phase of reduction (years to decades or more) reflecting longer
term changes in the rate of erosion/runoff loading to the waterbody and the potential buffering
effect of historical reservoirs of Hg in sediment. These findings suggest that we might not see an
equivalent fractional reduction in risk matching the reduction in aerial deposition for near-term
reductions (less than one year) due to historical reservoirs of Hg in sediment that continue to load
the benthic food web and buffer the Hg response. However, for reductions over a longer period
of time (sufficient for steady state, or near steady state conditions to be met in the aquatic
system), the proportionality assumption likely holds and we would expect to see a reduction in
risk matching the fractional reduction in Hg deposition. Since we have stated that our risk
estimates are based on an assumption that steady state, or near steady state conditions are met,
regardless of how long that takes, then the temporal response is not a factor and we have
increased confidence in applying the proportionality assumption.
46
-------
1.4.6.3 CMAQ mercury deposition modeling
We modeled total annual Hg deposition from U.S. and foreign anthropogenic and natural
sources as well as the fraction of Hg deposition from U.S. EGUs using the CMAQ model. The
Community Multi-scale Air Quality (CMAQ) model v4.7.1 (www.cmaq-model.org) is a state of
the science three-dimensional Eulerian "one-atmosphere" photochemical transport model used
to estimate air quality (Appel et al., 2008; Appel et al., 2007; Byun and Schere, 2006). CMAQ
simulates the formation and fate of photochemical oxidants, ozone, primary and secondary PM
concentrations, and air toxics over regional and urban spatial scales for given input sets of
meteorological conditions and emissions. Mercury oxidation pathways are represented for both
the gas and aqueous phases in addition to aqueous phase reduction reactions (Bullock and
Brehme, 2002). Mercury estimates from CMAQ have been compared to observations and other
mercury modeling systems in several peer reviewed publications (Bullock et al., 2008, 2009; Lin
et al., 2007). Additional information about the model, model inputs for this assessment, and
model evaluation are available in the Air Quality Modeling TSD (U.S. Environmental Protection
Agency, 2011).
The 36km and both 12km modeling domains were modeled for the entire year of 2005.
The emissions data used in the 2005 base year and 2016 total Hg emissions and U.S. EGU Hg
emissions zero-out cases are based on the 2005 v4.1 platform. Emissions are processed to
photochemical model inputs with the SMOKE emissions modeling system (Houyoux et al.,
2000). The 2016 total Hg emissions case is intended to represent the emissions associated with
growth and controls in that year projected from the 2005 simulation year. Only anthropogenic
emissions changed between the 2005 and 2016 simulations, all other model inputs are the same
in both simulations. Other North American emissions of criteria and toxic pollutants (including
mercury) are based on a 2006 Canadian inventory and 1999 Mexican inventory (U.S. EPA,
201 Ib).
Global emissions of criteria and toxic pollutants (including mercury) are included in the
modeling system through boundary condition inflow. The lateral boundary and initial species
concentrations are provided by a three-dimensional global atmospheric chemistry model, the
GEOS-CHEM model (standard version 7-04-11). The GEOS-CHEM predictions were used to
provide one-way dynamic boundary conditions at three-hour intervals and an initial
concentration field for the 36 km CMAQ simulations. The 36 km photochemical model
simulation is used to supply initial and hourly boundary concentrations to the 12 km domains.
Initial and boundary conditions for the projected future year (2016) 36 km simulations are the
same as the 2005 base year. The first 10 days of the 36 km modeling simulation are not used in
the analysis, which is beyond the number of days necessary to remove the influence of initial
conditions on mercury deposition estimates (Pongprueksa et al., 2008).
The boundary inflow for the CMAQ mercury modeling used in the national-scale
mercury risk assessment are based on a global model GEOS-CHEM simulation using a 2000
based global inventory as described in (Selin et al., 2007). A comparison of global mercury
emissions by continent for 2000 and 2006 was recently published in (Streets et al., 2009) and
show there is no discernable change in mercury emissions from Asia between 2000 and 2006.
Given these consistent emissions estimates from Asia, the 2005 boundary inflow to the 36 km
CMAQ domain was not adjusted. Mercury boundary conditions are the same for both the 2005
47
-------
and 2016 simulations based on the consistency in Asian mercury emissions between 2000 and
2006, the declining ambient mercury concentrations in the northern hemisphere since 2000
(Slemr et al, 2011), and the large uncertainties surrounding projected global inventories of
mercury emissions.
1.5 Differences between the 2005 Section 112(n) Revision Rule analysis and the
current analysis in support of the Appropriate and Necessary Finding
In 2005, EPA conducted a set of technical analyses to support revision of the 2000
appropriate and necessary finding.42 This section identifies key differences between the
watershed-level risk assessment completed in support of the 2005 revision rule and the current
risk assessment. These differences include both technical factors related to the design of the
assessments, as well as differences in the interpretation of potential public health significance of
the risk estimates generated. Key differences between the two analyses include:
Higher spatial resolution through use of CMAQ 12km grid cells; In this analysis, we modeled
Hg deposition using CMAQ at a 12 km grid resolution, whereas in the 2005 analysis, we used
CMAQ modeling with a 36 km grid cell resolution. The more refined spatial resolution at 12 km
is more appropriate for representing areas of elevated U.S. EGU-attributable deposition (and
total Hg deposition in general) compared with the 36 km resolution used in the 2005 analysis.
The 12 km resolution also matches up with the more refined HUC12 watersheds now being used
in the analysis, thereby allowing a more refined treatment of the intersection of aerial Hg
deposition and measured fish tissue concentrations at the watershed level.
Application of more refined HUC12 watersheds: The current analysis uses HUC12 watersheds as
the basis for risk estimation (these watersheds typically are 5-10 km on a side). By contrast, the
2005 analysis used HUCSs, which are much larger (averaging 40km on a side). The use of more
spatially refined watersheds increases the potential for capturing areas of elevated aerial Hg
deposition (combined with measured fish tissue levels).
Inclusion of updated fish tissue data: For this analysis, we included measured fish tissue data
collected between 2000 and 2010. By contrast, the 2005 analysis used data collected between
1999 and 2003, which was the best available data at the time.
Subsistence fisher activity better defined and considered more ubiquitous: Based on an extensive
review of available literature, we identified studies characterizing high-end self-caught fish
consumption for a wide variety of source populations (e.g., Hispanic, Vietnamese, Whites and
Blacks in the southeast, Great Lakes Tribal populations). Although in many cases, it was
necessary to extrapolate high-end fishing activity to regions beyond those covered in the
underlying studies, we believe that the literature supports the plausibility of high-end
subsistence-like fishing activity across the watersheds included in the analysis. Additionally, the
variety of studies identifying self-caught fishing activity at subsistence levels (i.e., a meal every
few days to a meal every day) for a variety of diverse SES-differentiated populations in different
42
U.S.EPA. 2005. Technical Support Document: Methodology Used to Generate Deposition, Fish Tissue
Methylmercury Concentrations, and Exposure for Determining Effectiveness of Utility Emission Controls.
48
-------
regions of the country, supports assessing this type subsistence fish consumption behavior across
the modeled watersheds.
By contrast, in the 2005 analysis, we concluded that the study data characterizing fishing activity
available at that time was limited in its ability to support modeling of subsistence fisher activity
for the following reasons: (a) it characterized regional or local activity that could not be readily
extrapolated more broadly, (b) fishing activity queried included consumption of saltwater
species, or (c) specific high-end percentiles were not identified (or if they were, they only
applied during specific harvesting periods - e.g., spearfishing months for Great Lakes Tribes).
Therefore, in the 2005 analysis, we applied a high-end self-caught percentile values (95th and
99th percentiles) based on Tribal fishing practices in the Northwest to watersheds across the
country.43 The updated literature review, for the current analysis, led us to revise several of our
earlier conclusions regarding high-end fishing activity. Specifically, while many of the studies of
subsistence-like activity are regional in nature, when considered together, we now conclude that
they support modeling subsistence-like fishing activity more broadly across the entire study area.
Additionally, while some of the studies may include saltwater fishing in addition to freshwater
(e.g., Burger, 2002), when those studies clearly covered both saltwater and freshwater self-
caught fish consumption, we concluded that it was reasonable to assume that subsistence-like
fishing activity could occur both at the coast and inland at freshwater bodies.44
For the current analysis, we used the 75th percentile fish tissue Hg concentration reflecting the
potential for high-end subsistence fishers to target larger fish, which would have greater
bioaccumulation potential relative to the average fish. By contrast, in the 2005 risk assessment,
we used the maximum of the average fish tissue Hg concentrations across species offish in a
given HUC8, which is potentially a more conservative approach (i.e., resulting in higher risk,
other factors equal).
Calculation of RfD-based HQ estimates including total and U.S. EGU-attributable risk and
calculation of IQ loss: For this analysis, we compared total exposure to the MeHg RfD to
generate an HQ estimate based on total Hg exposure for fishers at a given watershed.
Furthermore, to focus on the U.S. EGU component of that total risk, we have generated two
related risk metrics: (a) U.S. EGU incremental contribution to total risk which essentially
considers the magnitude of the HQ when deposition from U.S. EGUs is considered before taking
into account deposition and exposures resulting from other sources of Hg and (b) the percent of
43 These NW Tribal fishing estimates are subject to considerable uncertainty when extrapolated to other areas in the
U.S. These specific high-end fish consumption rates were derived for Tribes active in the Northwest who engage in
specific cultural practices focused around salmon fishing. There is also significant uncertainty in extrapolating this
type of highly-specific cultural-based fishing activity to other Tribes, let alone to other fish consumers in the U.S.
By contrast, extrapolation of more generalized high-consuming rates (for ethnic groups and Whites and Blacks) to
cover portions of the U.S. as was done in the current analysis is subject to less uncertainty given that these SES-
differentiated populations are defined more generally and therefore are likely to demonstrate more consistency in
culturally-related practices (such as subsistence fishing) across the country.
44 In the situation where a study specifically characterized low income high-end fishing populations, as is done in the
Burger 2010 study of activity in SC, we considered it reasonable to assume that low income individuals would likely
conduct their frequent fishing activity near home. In that case, some of these high-end fishers would likely be
located near the coast and some inland. In the case of subsistence-like fishing activity in the southeast, other studies
from rivers in that area also showed subsistence-like fish consumption rates when only freshwater rivers were
considered (e.g., Burger et al., 1999 focusing on fishing activity on the Savannah river in GA).
49
-------
total HQ risk attributable to U.S. EGUs. The calculation of U.S. EGU incremental contribution
to total HQ is identical to the IDI (index of daily intake) metric used in the 2005 analysis.
However an important distinction is that in the current analysis, we highlight the fact that this
U.S. EGU-related risk is always associated with a total HQ which is generally substantially
larger (i.e., the U.S.-EGU-attributable HQ should not be considered in isolation as was done in
the 2005 analysis with the IDI). By contrast, for the 2005 analysis both of the risk metrics used
(i.e., the IDI and the comparison of U.S. EGU-related fish tissue concentrations against EPA's
water quality criterion expressed as a Hg fish tissue value) essentially considered the U.S. EGU
portion of risk in isolation. These risk metrics in the 2005 analysis were not contrasted with the
much larger fraction of total Hg-related risk associated with the non-U.S. EGU portion of risk.
1.6 Detailed Example Calculation (watershed-level risk HQ)
This section provides a sample calculation of a watershed-level RfD-based HQ risk
estimate generated for the 2016 scenario and walks the reader through each of the calculation
steps associated with generating the estimate (see Figure 1-9). We cross-reference to sections of
the Revised TSD that cover each of the intermediate calculations and data inputs involved in
generating the risk estimate.45 Each of the calculation steps is summarized below:
• Step 1: Based on the fish tissue Hg measurements available for a given watershed (after
-th
filtering), generate the 75 percentile fish tissue Hg concentration. We provide the
median, 75th and 90th percentile values in Fig
values for this watershed), (see section 1.4.2).
median, 75th and 90th percentile values in Figure 1-9 to illustrate the spread in percentile
• Step 2: Obtain CMAQ Hg deposition estimates for the watershed for the 2005 and 2016
scenarios. Compute the ratio of these two factors (i.e., 2016 deposition/2005 deposition).
(see section 1.4.6.3).
• Step 3: Multiply the ratio from Step 2 by the fish tissue Hg concentration to project the
fish tissue Hg concentration value to represent the 2016 scenario for this watershed. This
step involves application of the proportionality assumption (see section 1.4.2.1).
• Step 4: Calculate intake rate for total Hg for the subsistence fisher scenario being
assessed in that watershed using equation with fish tissue Hg concentrations calculated in
Step 3(see section 1.4.4 for a discussion of this exposure calculation).
• Step 5: Compare the exposure estimate generated in step 4 to the MeHg RfD to generate
a hazard quotient reflecting total Hg exposure for the subsistence fisher at that watershed
(see section 1.4.5).
45 The SAB recommended including a cross reference between the calculation steps outlined in Figure 1-9 and the
discussion of variability and uncertainty presented in section 2-7. However, the complex interplay of sources of
uncertainty and variability with the analytical structure used in the risk assessment precludes presentation of a clean
linkage between sources of uncertainty and variability and the analytical framework used in the risk assessment.
Most sources of variability and uncertainty impact either directly or indirectly multiple (if not all) of the calculations
steps involved in generating risk estimates. Careful consideration of the analytical approach reflected in Figure 2-1
and the flow diagram outlining major analytical steps of the risk assessment presented in Figure 1-2 should allow the
reader to determine which elements of the risk model are impacted by specific sources of uncertainty and/or
variability.
50
-------
• Step 6: Obtain CMAQ Hg deposition estimates for the watershed for both total and U.S.
EGU-attributable Hg for the 2016 scenario. Compute the ratio of these two factors (i.e.,
U.S. EGU/total deposition). This ratio will be used to estimate the fraction of total risk
attributable to U.S. EGUs (see section 1.4.6).
• Step 7: Estimate the fraction of the total HQ risk that is attributable to U.S. EGUs (for the
2016 scenario). This estimate is based on the proportionality assumption and uses the
ratio developed in step 6 (see section 1.4.6).
51
-------
#: calculation step
Example-Lower Kettle Creek in north central Pennsylvania (HUC12 ID: 20502030106)
Fish tissue (ppml
Median: 0.24
75th: 0.32
90th: 0.34
6 samples
Mercury 95th%
conversion consumption
factor: rate:
0.95 173g/day
Total Hg dep
over watershed
(CMAQ):
2005 = 30.42
2016 = 25.19
Or a ~17% reduction
IR = FTC* MCF* FCR* CAP
BW
PROJECTED
Fish tissue
(ppml 2016
Mean: 0.20
75th: 0.26
90th: 0.28
IR: daily MeHg intake rate (ug/kg-day)
FTC: MeHg fish tissue concentration (ug/g or ppm)
MCR: mercury conversion factor -fraction fish tissue
mercury that is MeHg (unitless)
FCR: fish consumption rate (g/day)
FPCAF: fish preparation/cooking adjustment factor
RfD: methylmercury RfD
FracEGU: fraction dep from US ECU
Fish
preparation/
cooking
adjustment
Factor:
1.5
IR= 1.01 ug/kg-day
Total HQ= JR_= 1.01 = 10.1
RfD 0.1
MeHg RfD:
0.1 ug/kg-day
\7
2016 Hg dep
over watershed
(CMAQ):
total = 25.19
US ECU = 1.26
US ECU is 5%
FracEGU = 0.050
US ECU risk = HQ * FracEGU
= 10.1 *0.050 = 0.504 (i.e., <1)
2
Figure 1-9 Sample Calculation for watershed-level Risk HQ
52
-------
2 Discussion of Analytical Results
This chapter provides a discussion of the results of the 2016 simulation, including both
risk estimates as well as intermediate calculations and inputs associated with generating those
risk estimates. EPA's projection of total Hg emissions from U.S. EGUs in 2016 (after other
CAA-related regulations are fully implemented) is 29 tons.46 This national estimate is the same
as the estimate of recent U.S. EGU Hg emissions based on information collected from industry
through the Information Collection Request (ICR). This shows that significant reduction in U.S.
EGU Hg emissions has occurred since 2005, when Hg emissions were estimated to be 52.9 tons,
but that few additional reductions are expected to occur without additional regulations to reduce
mercury emissions. The reductions between 2005 and 2016 are largely due to regulations and
federal enforcement actions that achieve Hg reductions as a co-benefit of controls for NOx and
SC>2 emissions. Given these estimates of total Hg emissions, characterization of "current
conditions" is better represented by the 2016 Scenario than the 2005 Base Case, since total Hg
emissions for the former is equal to our projection of recent emissions. By contrast, the 2005
analysis reflects total Hg emissions (52.9 tons) which are significantly higher than our estimate
of recent emissions. For this reason, as mentioned earlier, we only present risk estimates for the
2016 Scenario and have not generated risk estimates for the 2005 scenario (although the 2005 Hg
deposition estimates are used in scaling fish tissue Hg concentrations to represent concentrations
in 2016).
In this section, we provide a brief overview of critical design elements of the risk
analysis that the reader should keep in mind when reviewing the results (section 2.1). We then
discuss the intermediate inputs and outputs for individual analytical steps associated with the risk
assessment including: (a) Hg deposition from U.S. EGUs as modeled using CMAQ (section 2.2),
(b) fish tissue Hg concentrations (section 2.3), (c) relationship between Hg deposition and Hg
fish tissue concentrations (section 2.4) and (d) the MeHg RfD-based HQ risk assessment results
for the 2016 scenario (section 2.5). In discussing each category of results, emphasis is placed on
identifying key policy-relevant observations. In section 2.6, we discuss the results of several
sensitivity analyses conducted to characterize the potential impact of specific sources of
uncertainty on the risk estimates. In section 2.7, we discuss variability and uncertainty related to
the risk assessment. In Section 3, we provide a summary of critical observations from the
analysis, which draws on information provided in sections 2.2 through 2.7.
2.1 Key design elements to consider when reviewing the risk assessment results
The following design elements of the analysis should be considered when reviewing the
results:
• The analysis focuses on subsistence-like fishing activity at inland freshwater bodies. The
analysis is not intended to capture more generalized recreational fishing activity or to
reflect self-caught fisher exposure associated with saltwater fishing or fishing in the Great
46 The modeling conducted for the March risk assessment included a representation of the proposed Cross State Air
Pollution Rule, which was finalized in June 2011. We have evaluated differences in projected Hg emissions after
taking into account differences between the proposed and final CSAPR and determined that the difference is less
than 2 tons nationally, which would have little expected impact on our risk estimates.
53
-------
Lakes. In comparing any risk profiles generated in this analysis to risks estimated in other
contexts, the specific focus on these high-end populations shows risks substantially
higher than risks for recreational fishers.
• The risk estimates generated are based on a set of female subsistence fish consumer
scenarios evaluated at watersheds where (a) we have fish tissue Hg data and (b) this type
of subsistence fisher activity could potentially occur. Therefore, the risk assessment is
watershed-focused, does not provide population-representative risk estimates for
subsistence fishers.
• Risks are estimated for 3,141 watersheds. This watershed coverage (only about 4% of
U.S. watersheds) leaves much of the country not covered by the analysis, including a
substantial number of watersheds with relatively elevated levels of U.S. EGU-related Hg
deposition. Further, the eastern part of the U.S. is more heavily represented in the
watershed-level estimates of risk. Given that U.S. EGU Hg deposition is generally higher
in the eastern part of the U.S., the fact that the risk assessment is focused on this part of
the country is considered to be a strength of the analysis.
• The analysis uses a proportionality assumption to link changes in Hg deposition (over
watersheds) to changes in Hg fish tissue levels. This approach assumes that near steady-
state conditions are met, which may take years to decades at a given watershed following
changes in Hg deposition.
• The analysis estimates risk based on MeHg RfD-based HQ. The U.S. EGU-attributable
HQ should always be considered in the context of total HQ which is typically
substantially larger than the U.S. EGU-attributable HQ (reflecting the large contribution
from non-U.S. EGU sources of Hg). We also assessed IQ loss in children, however, due
to concerns that the IQ loss endpoint may not fully capture all of the neurodevelopmental
effects associated with MeHg exposure, we have moved discussion of the IQ loss
estimates to Appendices A and B.
• Because it is not feasible to enumerate the female subsistence fish consumers modeled in
this analysis, we could not generate distributions of population-weighted risk for specific
scenarios assessed (e.g., low income Hispanic fishers, or Tribal fishers in the vicinity of
the Great Lakes). However, we do believe, based on surveys of their behavior, that this
type of subsistence-like activity could reasonably be expected to occur across some
fraction of the 3,141 watersheds included in the analysis. Therefore, we have assessed
female subsistence fish consumer risk for each watershed where we have fish tissue Hg
data. We then consider the fraction of watersheds that meet the risk characterization
criteria outlined in the risk characterization framework (see Section 1.3).
2.2 Mercury Deposition from U.S. EGUs as Modeled Using CMAQ
Below we provide the results of the CMAQ modeled Hg deposition for the 2005 and
2016 scenarios for total deposition and U.S. EGU-attributable deposition.47 For this Revised
47 During the Science Advisory Board (SAB) review of the March TSD held on June 15-17th, Panel members raised
questions regarding patterns of mercury deposition reflected in Figures 2-1 though 2-4 of the March version of the
54
-------
TSD we have also included maps of wet and dry deposition for the 2005 and 2016 scenarios.48
Unlike the results presented in the subsequent sections that are limited to watersheds with fish
tissue data, we modeled Hg deposition in all 88,000 watersheds in the continental U.S. We
present a set of bulleted observations at the end of the section that draws on information
conveyed in the figures and tables. The set of figures and tables presented include:
• Figure 2-1 and 2-2: Maps presenting CMAQ modeling results for total Hg deposition
(|ig/m2) at the watershed-level, for the 2005 and 2016 scenarios respectively.
• Figures 2-3 and 2-4: Maps presenting CMAQ modeling results for U.S. EGU-attributable
Hg deposition (|ig/m2) at the watershed-level, for the 2005 and 2016 scenarios,
respectively.
• Figures 2-5 through 2-8: Maps presenting CMAQ modeling results for wet and dry Hg
deposition (|ig/m2) for the 2005 and 2016 scenarios. These estimates are presented at the
original CMAQ 12km grid resolution and have not been interpolated to the watershed-
level. Maps presenting wet and dry deposition modeling results have been included to
allow readers to consider patterns of deposition over specific locations and potentially
compare those with measured data, in those instances where relevant data are (or
become) available.
• Table 2-1: Summary of statistics (mean, 50th, 75th, 90th, 95th and 99th percentiles) for total
Hg deposition and U.S. EGU-attributable deposition for the 2005 and 2016 scenarios.
• Table 2-2: Summary of statistics (mean, 50th, 75th, 90th, 95th and 99th percentiles) for U. S.
EGU-attributable deposition as a percent of total deposition for the 2005 and 2016
scenarios.
• Table 2-3: Summary of statistics (mean, 50th, 75th, 90th, 95th and 99th percentiles) for
percent reduction of (a) total Hg deposition, and (b) U.S. EGU-attributable deposition,
based on comparison of the 2016 scenario against the 2005 scenario.
Mercury Risk TSD. The figures in the March TSD were intended to show annual total mercury (Hg) deposition per
unit area (in units of ug/m2) by watershed, however they actually displayed intermediate calculations that had not
been adjusted by the waterbody-specific surface areas. We corrected the figures and provided them in a memo to the
docket (EPA-HQ-OAR-2009-0234-15522). The figures presented here are the corrected figures and match those
presented in that memo.
48 Inclusion of maps presenting wet and dry deposition modeling results reflects suggestions made by SAB Panel
members.
55
-------
2005 HUC-level total Hg deposition (ug/m2)
^B <16
^H 16-26
26-43
|^| 43 - 99
^H >gg
Figure 2-1 Total Mercury Deposition by HUC (ug/m ) for the 2005 Scenario
56
-------
2016 HUC-leve! total Hg deposition (ug/m2)
16-26
26-43
43 - "
>99
Figure 2-2 Total Mercury Deposition by HUC (ug/m ) for the 2016 Scenario
57
-------
2005 HUC-level EGU Hg deposition (ug/m2)
^B <0.18
^B 0.18-0.6
0.6 - 2.0
2.0- 10
Figure 2-3 U.S EGU-Attributable Mercury Deposition by HUC (jig/m ) for the 2005 Scenario
58
-------
2016 HUC-level ECU Hg deposition (ug/m2)
^B <0.18
^B 0.18-0.6
0.6-2.0
2.0 -10
Figure 2-4 U.S EGU-Attributable Mercury Deposition by HUC (ug/m2) for the 2016 Scenario
59
-------
2005 CMAQ 12km Hg Wet Deposition (ug/m2)
j^B<1Q
j^B 10-15
15-25
^B 25 ~ 50
^H >50
Figure 2-5 Mercury Wet Deposition by 12km Grid Cell (ug/m ) for the 2005 Scenario
60
-------
2005 CMAQ 12km Hg Dry Deposition (ug/m2)
^H <10
10-15
15-25
^B 25-5°
j^HJ >50
Figure 2-6 Mercury Dry Deposition by 12km Grid Cell (ug/m ) for the 2005 Scenario
61
-------
2016 CMAQ 12km Hg Wet Deposition (ug/m2)
10-15
15-25
25-50
>50
Figure 2-7 Mercury Wet Deposition by 12km Grid Cell (ug/m ) for the 2016 Scenario
62
-------
2016 CMAQ 12km Hg Dry Deposition (ug/m2)
^B <10
^B lo-15
15-25
| 25-50
• >50
Figure 2-8 Mercury Dry Deposition by 12km Grid Cell (jig/m ) for the 2016 Scenario
63
-------
Table 2-1 Comparison of total and U.S. EGU-attributable mercury deposition (ug/m ) for
the 2005 and 2016 scenarios *
Statistic
Mean
Median
75th percentile
90th percentile
95th percentile
99th percentile
2005 scenario
Total Hg
Deposition
19.41
17.25
23.69
30.78
36.85
58.32
U.S. EGU-attributable
Hg Deposition
0.89
0.24
1.07
2.38
3.60
7.77
2016 scenario
Total Hg
Deposition
18.66
16.59
22.83
29.90
35.16
56.23
U.S. EGU-attributable
Hg Deposition
0.34
0.15
0.46
0.85
1.18
2.41
* Values are based on CMAQ results interpolated to the watershed -level and reflect trends across all -88,000
watersheds in the continental U.S. Percentiles for total Hg deposition and U.S. EGU attributable Hg
deposition are not matched, e.g. the watershed with the 99th percentile for total Hg deposition will not be
the same watershed as the watershed with the 99th percentile for U.S. EGU attributable Hg deposition.
Table 2-2 Comparison of percent of total mercury deposition attributable to U.S. EGUs for
2005 and 2016.*
Statistic
Mean
Median
75th percentile
90th percentile
95th percentile
99th percentile
2005 scenario
5%
1%
6%
13%
18%
30%
2016 scenario
2%
1%
3%
5%
6%
11%
* Values are based on CMAQ results interpolated to the watershed -level and reflect trends across all -88,000
watersheds in the U.S.
Table 2-3 Comparison of percent reduction of total mercury deposition, and U.S. EGU-
attributable deposition, based on comparing the 2016 scenario against the 2005
scenario.*
Statistics
Mean
Median
75th percentile
90th percentile
95th percentile
99th percentile
Percent Change in Total
Hg Deposition
-4%
-1%
-5%
-12%
-16%
-27%
Percent Change in U.S.
EGU-attributable Hg
Deposition
NC**
-41%
-70%
-80%
-85%
-91%
* Values are based on CMAQ results interpolated to the watershed -level and reflect trends across all -88,000
watersheds in the U.S.
** A mean value was not calculated (NC) for this category due to presence of a number of watersheds with very
small U.S. EGU-attributable deposition values which skewed this distribution.
64
-------
We made the following observations based on the information presented in Figures 2-2
through 2-4 and in Tables 2-1 through 2-3 regarding estimates of total and U.S. EGU-attributable
Hg deposition for the 2005 and 2016 scenarios:
• Patterns of total and U.S. EGU-related Hg deposition differ considerably: Areas of
elevated total Hg deposition are distributed around the country (e.g., west coast, areas in
Nevada, southern Mississippi, West Virginia, southeastern Georgia) (see Figures 2-1 and
2-2). By contrast, U.S. EGU Hg deposition is concentrated in the eastern U.S., especially
in the Ohio River Valley (see Figures 2-3 and 2-4). Figures 2-3 and 2-4 also illustrate that
while some near-coastal areas and portions of the Great Lakes have elevated U.S. EGU-
attributable deposition, many of the highest areas (and largest expanses) of U.S. EGU-
attributable deposition occur inland (e.g., Ohio River Valley, areas in northeast Texas and
along the Mississippi River).
• U.S. Hg deposition is generally dominated by sources other than U.S. EGUs and the
contribution from U.S. EGUs decreases between the 2005 and 2016 scenarios: On
average., U.S. EGUs contribute 5% of total Hg deposition for the 2005 scenario, which
decreases to 2% for the 2016 scenario (see Table 2-2). The remaining Hg deposition (i.e.,
-95% and -98%, respectively for the two scenarios) originates from other U.S. sources
of Hg emissions and from international sources (both anthropogenic and natural). U.S.
EGU-attributable deposition decreases considerably between the 2005 and 2016
scenarios, primarily from implementation of the Cross State Air Pollution Rule (CSAPR),
state Hg regulations and Federal enforcement actions.49 The median reduction in U.S.
EGU-attributable deposition was 41% with reductions ranging up to 85% for the 95th
percentile watershed (see Tables 2-2 and 2-3).
• The contribution of U.S. EGU-attributable deposition to total deposition varies across
watersheds andean represent a relatively large fraction in some instances: In the 2005
scenario, while on average, U.S. EGUs only contributed 5% of total Hg deposition in the
U.S., this contribution ranged up to 30% for the 99th percentile watershed (see Table 2-2).
While overall U.S. EGU -attributable deposition decreased substantially between the
2005 scenario and the 2016 scenario, U.S. EGUs contributed 11% of total Hg deposition
for the 99th percentile watershed in 2016 ((see Table 2-2).
2.3 Fish Tissue Mercury Concentrations
This section characterizes the subset of U.S. watersheds with fish tissue Hg data included
in the current the risk assessment. This dataset includes (a) the original HUC-level fish tissue
dataset used in the March version of the risk assessment (post augmentation data set with 2,317
watersheds) and (b) the additional 940 watersheds with fish tissue data identified as part of our
refinement to the March risk assessment (for a total of 3,141 watersheds). In this section, we
provide the 75th percentile fish tissue sample for each watershed. As discussed in Sections
1.4.2.1 and 1.4.6, we used the proportionality assumption together with Hg deposition coverages
49 Controls on PM precursors, including directly emitted PM and SO2, can significantly reduce divalent and particle-
bound mercury, both of which primarily deposit locally and regionally. For more information on the emission
reductions from CSAPR, see the final Regulatory Impact Analysis, which is available at
http://www.epa.gov/airtransport/pdfs/FinalRIA.pdf.
65
-------
for 2005 (total and U.S. EGU-related) and 2016 (total and U.S. EGU-related) to project fish
tissue Hg concentrations for the 2016 scenario and to estimate the U.S. EGU-attributable
fraction of total fish Hg.
We provide figures and tables to summarize the fish tissue data. As discussed in Section
2.2, most of the areas with elevated U.S. EGU-attributable Hg deposition are located in the
eastern U.S., where we also have more Hg fish tissue data and most of the U.S. EGU emission
reductions between 2005 and 2016. We also present a set of bulleted observations based on the
figures and tables. The set of figures and tables presented include:
• Figure 2-9: Map of 3,141 watersheds with fish tissue sampling data used in the current
version of the risk assessment, including (a) watersheds included in the previous version
of the risk assessment (2010 MFT) and (b) additional fish tissue data incorporated to
enhance the current version of the risk assessment (augmentation). This map illustrates
the uneven coverage of the fish tissue dataset across the continental U.S and highlights
which states were included in the additional data incorporated for this version of the risk
assessment (i.e., the augmentation HUCs shown in red).
• Figure 2-10: Map of 3,141 watersheds with fish tissue data showing distribution of
sampling frequency (number offish tissue measurements) across watersheds. This map
can be used to show the relationship between trends in sampling frequency and fish tissue
Hg concentrations.
• Figure 2-11 and 2-12: Maps presenting Total Fish Tissue Mercury Concentrations for the
2005 and 2016 scenarios respectively.
• Figures 2-13 and 2-14: Maps presenting EGU-Attributable Fish Tissue Mercury
Concentrations for 2005 and 2016 scenarios, respectively.
50
• Table 2-4: Comparison of watershed-level fish tissue Hg concentrations (including means
and various percentiles) for the March and revised version of the national-scale Hg risk
assessment. Table 2-4 includes percentile data for (a) 2010 MFT dataset (the fish tissue
data used in the March risk assessment), (b) the augmentation dataset (additional fish
tissue data collected to supplement data used in the last risk assessment) and (c) the 2011
MFT dataset (the dataset used in the current revised version of the risk assessment).
• Table 2-5: Summary of statistics (mean and various percentiles) for both total and U.S.
EGU-attributable Hg fish tissue levels (for the 2005 and 2016 scenarios). These statistics
are based on watershed-level data. In addition, this table also presents the percent
reduction (between the 2005 and 2016 scenarios) for both total and U.S. EGU-
attributable Hg fish tissue levels.
50 Note, in the March version of the Mercury Risk TSD, we included a series of maps showing fish tissue
Hg concentrations at the watershed-level for the upper 10 percent of watersheds with regard to total fish
tissue Hg concentrations and EGU-attributable fish tissue Hg concentrations. However, we decided to
remove these maps from the revised version of the Mercury Risk TSD given the limited coverage that the
fish tissue Hg sampling data provides for watersheds across the U.S.
66
-------
HUCs with fish tissue mercury data
(pre and post augmentation)
3.141 of -88.000 HUC12s with fish tissue data
^H 2010 MFTdataset
I augmentation dataset
Figure 2-9 Set of 3,141 HUC12s with Fish Tissue Mercury Data Used in the Risk Assessment*
* Includes 2010 Mercury Fish Tissue (MFT) dataset used in the March version of the risk assessment as well as the
augmentation dataset incorporated for the current version of the risk assessment.
67
-------
Fish Tissue Sampling Frequency
by HUC for the Set of HUCs
with Fish Tissue Data
o 1
* 2-3
* 4-8
* 9-18
* 19-360
Figure 2-10 Fish Tissue Measurement Sampling Frequency for HUCs with Fish Tissue Data Included in the Risk Assessment
68
-------
Fish Tissue Mercury Data
(HUC12-level 75th percentile values, ppm)
* 0.000000-0.109143
o 0.109144-0.195000
o 0.195001 -0.300000
o 0.300001 -0.479000
* 0.479001 -6.605000
th
Figure 2-11 Total Fish Tissue Mercury Concentrations for 2005 Scenario (HUC12-level 75 percentile values, ppm)
69
-------
Fish Tissue Mercury Data Projected for 2016
(HUC12-level 75th percentile values, ppm)
* 0.000000-0.109143
o 0.109144-0.195000
o 0.195001 -0.300000
* 0.300001 -0.479000
* 0.479001 -6.605000
-th
Figure 2-12 Total Fish Tissue Mercury Concentrations Projected for 2016 Scenario (HUC12-level 75 percentile values,
ppm)*
* Values estimated by adjusting fish tissue Hg data presented in Figure 2-12 to reflect different patterns of Hg deposition in
2016 (relative to 2005).
70
-------
ECU Fish Tissue Mercury Data
(HUC12-level 75th percentile values, ppm)
* O.DOOOOO - 0.004323
o O.DD4324-0.010451
o 0.010452-0.019788
o 0.019789-0.037493
* 0.037494 - 0.408582
Figure 2-13 EGU-Attributable Fish Tissue Mercury Concentrations for 2005 Scenario
(HUC12-level 75th percentile values, ppm)*
* Scales are different between the EGU-attributable and total Hg fish tissue concentrations maps.
71
-------
EGU Fish Tissue Mercury Data
Projected for 2016
(HUC12-level 75th percentile values, ppm)
* O.ODODDO - 0.004323
o 0.004324-0.010451
o 0.010452-0.0197B8
o 0.019789-0.037493
* 0.037494-0.4085S2
Figure 2-14 EGU-Attributable Fish Tissue Mercury Concentrations Projected for 2016 Scenario
(HUC12-level 75th percentile values, ppm)*
* Scales differ between the EGU-attributable and total Hg fish tissue concentrations maps.
72
-------
Table 2-4 Comparison of HUC-Level Fish Tissue Mercury Concentrations Across Datasets
Used in the March Version and Current Version of the Risk Assessment
statistic
Mean
Median
75th %
90th %
95th %
99th %
Hg fish tissue concentration (ppm)
(HUC12-level 75th percentile)
2010 MFT dataset
(used in the March
version of the risk
assessment)
(2,317)
0.31
0.23
0.39
0.67
0.91
1.33
Augmentation
dataset
(collected as part of
refining analysis)
(940)
0.37
0.30
0.47
0.72
0.86
1.44
2011 MFT dataset
(used in the current
version of the risk
assessment)
(3,141)
0.32
0.25
0.42
0.67
0.89
1.35
Table 2-5 Comparison of total and U.S. EGU-attributable Hg fish tissue concentrations
(including % change) for the original fish tissue dataset and 2016 scenarios*
Statistic
Mean
Median
75th %
90th %
95th %
99th %
Hg fish tissue concentration (ppm)
Used for 2005 scenario
Total
0.32
0.25
0.42
0.67
0.89
1.35
U.S. EGU-
attributable
0.024
0.014
0.032
0.057
0.081
0.152
U.S. ECU
as percent
of total
9.1%
6.0%
13.5%
20.4%
26.2%
42.9%
Projected for 2016 scenario
Total
0.30
0.22
0.40
0.65
0.86
1.30
U.S. EGU-
attributable
0.008
0.005
0.011
0.018
0.026
0.045
U.S. ECU
as percent
of total
3.4%
2.6%
4.4%
6.6%
8.7%
16.4%
% change (2016 versus
original fish tissue
dataset) in Hg fish
tissue concentration
Total
-8.0%
-4.9%
-12.7%
-18.6%
-23.0%
-38.3%
U.S. EGU-
attributable
-43.9%
-61.4%
-76.0%
-82.6%
-85.4%
-91.7%
* These percentiles are not for matched HUCs (i.e., the 99* P6™"1*16 for total Hg fish concentration for the 2005
scenario could occur in a different HUC than the 9^Pe'ccentlle U.S. ECU attributable percent reduction).
We made the following observations based on Figures 2-10 through 2-14 and in Tables 2-
4 and 2-5:
• U.S. EGU-attributable fish tissue Hg levels are higher in the eastern half of the U.S.:
This reflects primarily the fact that levels of U.S. EGU Hg deposition (that largely drives
U.S. EGU-attributable Hg fish tissue levels) are much higher in the east (see Figures 2-3
and 2-4).
• Augmenting the fish tissue Hg dataset significantly improved coverage for Wisconsin,
Minnesota, Pennsylvania and New Jersey: Although the additional fish tissue Hg data
73
-------
significantly expanded coverage for watersheds in areas with elevated U.S. EGU-
attributable Hg deposition (Pennsylvania) and for areas with relatively elevated fish
tissue Hg concentrations (Minnesota and Wisconsin), the additional data did not
substantially change the overall distribution offish tissue Hg concentrations. This
conclusion is based on a comparison of percentiles across the different datasets (i.e., 2010
MFT, augmentation, and 2011 MFT datasets) as presented in Table 2-4.
• U.S. EGUs contribute a larger fraction to total Hg fish tissue levels in the U.S. than they
do to total Hg deposition (in terms of percent), reflecting the fact that Hgfish tissue
samples are focused in the east where U.S. EGU-attributable deposition is greater'. While
U.S. EGUs contribute -5% of total Hg deposition in the U.S. (for the 2005 scenario - see
Table 2-2), their contribution to Hg fish tissue levels (summarized at the watershed-level)
for the 2005 scenario is larger at -9% (see Table 2-5). This reflects the fact that Hg fish
tissue samples are heavily weighted in the eastern portion of the U.S. where U.S. EGU
Hg deposition is typical higher than in the west. By providing greater coverage for the
eastern half of the country, the Hg fish tissue sampling data generally provides greater
coverage for regions with potentially greater U.S. EGU-attributable risk.
• Relative to the combined impact of other sources, U.S. EGUs represent a smaller, but
still potentially important contributor to total fish tissue Hg concentrations: U.S. EGUs
contribute -9% of Hg fish tissue levels on average under the 2005 scenario (see Table 2-
5). Under the 2016 scenario, the U.S. EGU contribution decreases to ~ 3% on average
(see Table 2-5). While U.S. EGU-attributable Hg fish tissue decreases notably between
the 2005 and 2016 scenarios, the impact on total Hg fish tissue levels is not that
noticeable given that U.S. EGUs contribute a relatively small fraction on total Hg fish
tissue levels in general (contrast the pattern of reduction seen in Figures 2-13 and 2-14
for U.S. EGU-attributable Hg fish tissue levels with the relatively smaller changes seen in
Figures 2-11 and 2-12 for total Hg fish tissue levels).
• Despite the relatively small fraction of U.S. EGU-attributable fish tissue Hg on average,
for a subset of watersheds, U.S. EGU-attributable deposition has a significant
contribution on fish tissue: Under the 2005 scenario, U.S. EGUs contribute up to 43% of
total Hg fish tissue levels (for the 99th percentile watershed). Under the 2016 Scenario,
this pattern is reduced, but U.S. EGUs can still contribute up to 16% of total fish tissue
Hg concentrations (again, for the 99th percentile watershed) (see Table 2-5).
2.4 Comparing Patterns of Hg Deposition with Hg Fish Tissue Data for the 3,141
Watersheds Included in the Risk Assessment
We also compared spatial patterns of CMAQ modeled Hg deposition and fish tissue Hg
concentrations, which can inform interpretation of the risk estimates. Specifically we can
consider: (a) whether the watershed-level Hg fish tissue levels are correlated with total Hg
deposition, (b) how patterns of Hg deposition for the 3,141 watersheds with Hg fish tissue data
compare with patterns for the 88,000 watersheds in the continental U.S. and (c) to what extent
the watersheds for which we have fish tissue Hg data correspond to areas with elevated U.S.
74
-------
EGU-attributable deposition. To address these questions, we developed a series of figures
including:
• Figures 2-15 and 2-16: Maps comparing (a) watersheds with fish tissue Hg data with (b)
the general pattern of U.S. EGU-attributable Hg deposition across the continental U.S.
(for 2005 and 2016). These maps include both the fish tissue dataset used in the March
TSD (2010 MFT dataset) and the augmented dataset. These maps show the degree of
overlap between fish tissue Hg data and areas of elevated U.S. EGU-attributable Hg
deposition (for both 2005 and 2016 scenarios).51
• Figure 2-17: Plot comparing the fish tissue Hg concentrations versus total Hg deposition
(2005 scenario) by watershed. This plot allows consideration for whether there appears to
be a correlation between these two factors at the watershed level.
• Figure 2-18: Cumulative distribution plots comparing U.S. EGU-attributable deposition
for the 3,141 watersheds used in the risk assessment with U.S. EGU-attributable
deposition of the 88,000 watersheds in the continental U.S. Given emphasis in the risk
assessment on the 2016 scenario, we only provided plots for this scenario. These plots
allow us to consider whether the watersheds with fish tissue Hg data tended to fall in
regions with higher U.S. EGU-attributable Hg deposition and the degree to which this
subset of watersheds provided coverage for areas with relatively elevated U.S. EGU Hg
deposition across the country.
51 These two maps replace Figures 2-15 and 2-16 from the March TSD. Those earlier maps also compared
watersheds with fish tissue Hg data to areas of the country that had elevated U.S. EGU-attributable Hg deposition.
However, in identifying areas of elevated U.S. EGU-attributable Hg deposition for those earlier maps, we attempted
to be much more precise. Specifically, we used a deposition threshold for identifying areas of high U.S. EGU-
attributable Hg deposition based on the average rate of deposition over watersheds modeled for risk that had U.S.
EGU-attributable exposure exceeding the MeHg RfD. In other words, we identified a deposition level associated
generally with watersheds having high U.S. EGU-attributable risks and used this bright line as the basis for
identifying areas of high U.S. EGU deposition across the U.S.; the implication being, that these could be areas of
high U.S. EGU-attributable risk. However, elevated U.S. EGU-attributable risk typically reflects both elevated U.S.
EGU-attributable Hg deposition and elevated baseline fish tissue Hg concentrations. Therefore, attempting to
consider only U.S. EGU-attributable deposition as a means for identifying areas of potential elevated risk is subject
to uncertainty, since it does not also consider the baseline fish tissue Hg concentrations. The current set of figures
seeks to support a more general comparison of the watersheds we have modeled for risk against the national pattern
of U.S. EGU-attributable Hg deposition, by not attempting to identify a bright line for classifying areas of the
country with regard to the magnitude of elevated U.S. EGU-attributable Hg deposition.
75
-------
Comparison of watersheds
with Hg fish tissue data and
pattern of U.S. ECU-attributable
Hg deposition (2005)
^B Augmentation fish tissue dataset
|^| 2010 MFT fish tissue dataset
Hg Deposition |ug m2)
0.000000-0.180000
^•0.180001 -o.eooooo
IH 0.600001 -2.000000
1^2.000001 -10.000000
^M 10.000001 -99.484081
Figure 2-15 Comparison of Locations of Watersheds with Fish Tissue Hg Data* with Pattern of U.S. EGU-attributable Hg
Deposition (2005 scenario)
* Fish tissue Hg data includes (a) 2010 MFT data used in March version of the risk assessment and (b) augmentation data
added for this revised version.
76
-------
Comparison of watersheds
with Hg fish tissue data and
pattern of U.S. ECU-attributable
Hg deposition (2016)
m Augmentation fish tissue dataset
^|20IO MFT fish tissue dataset
Hg Deposition (ug/m2)
0.000000-0.180000
IH 0.180001 -o.eooooo
O.S00001 -2.000000
^H 2.000001 -10.000000
^•10.000001 -31.767133
Figure 2-16 Comparison of Locations of Watersheds with Fish Tissue Hg Data* with Pattern of U.S. EGU-attributable Hg
Deposition (2016 scenario)
* Fish tissue Hg data includes (a) 2010 MFT data used in March version of the risk assessment and (b) augmentation data
added for this revised version.
77
-------
0.2
0.4
0.6
0.8
1.2
1.4
1.6
1.8
HUC-level fish tissue mercury levels (ppm)
for 3,141 HUCs used in risk assessment
Figure 2-17 For the 2005 Scenario, Scatter Plot of Hg Fish Tissue Concentrations Versus
Total Hg Deposition for the 3,141 Watersheds Included in the Risk Assessment
Cumulative Distribution of Mercury Deposition by Watershed
I
1
1
+ e a ic
Hj deposition (i«/ni2|
-20U HUCs-.vitti fish dab EfflJ Hfedsp — ALJLHUCs 1016 E6U Dep
Figure 2-18 Cumulative distribution plots of U.S. EGU-attributable Hg deposition over the
3,141 watersheds used in modeling the high-end female consumer population as
contrasted with all 88,000 watersheds (2016 Scenario).
78
-------
We made the following observations based on the information presented in Figures 2-15
through 2-18 regarding how estimates of Hg deposition estimates relate to measured fish tissue
Hg concentrations at the watershed-level:
• The fish tissue Hg sampling data (watersheds modeled for risk) provide varying coverage
for areas with elevated U.S. EGU-attributable Hg deposition. The augmented dataset
significantly expanded coverage in key areas. However, the number of "at-risk"
watersheds may still be substantially higher than we estimated: As depicted in Figures 2-
15 and 2-16, the 3,141 watersheds used in the risk assessment provide reasonable
coverage for some of the regions having the highest U.S. EGU-attributable Hg deposition
(i.e., Ohio River Valley), especially with the augmented dataset for Pennsylvania.
However, even with the expanded coverage, the majority of areas with elevated U.S.
EGU-attributable deposition are not covered in the risk assessment. Therefore, we
believe that the number of watersheds with elevated U.S. EGU-attributable exposure
and/or risk could be substantially larger (depending on the underlying fish tissue Hg
concentrations).52
• Hgfish tissue levels are not correlated with total Hg deposition because the relationship
is highly dependent on methylation potential of individual waterbodies: As shown in
Figure 2-17, total Hg fish tissue levels at the watershed-level are not correlated with
levels of total Hg deposition across watersheds (i.e., the highest total Hg deposition
watersheds do not always have the highest fish tissue Hg concentrations). This is not
unexpected because the relationship between total Hg deposition and total Hg fish tissue
levels is highly dependent on the methylation potential at the waterbody, which is driven
in part by the presence of wetlands, levels of aqueous organic carbon, Ph and sulfate
deposition (see section 1.4.6).
• Hgfish tissue samples were generally collected in regions with elevated total Hg
deposition: As demonstrated in Figure 2-18, Hg fish tissue sampling appears to have
favored areas with relatively higher total Hg deposition.53 This can be seen by comparing
cumulative plots of modeled watersheds (where we have fish tissue Hg data) against plots
for the 88,000 watersheds. This comparison suggests that watersheds where fish tissue
Hg data were collected tended to have higher total Hg deposition than the full set of
watersheds. This likely reflects to some extent, the fact that fish tissue sample are focused
in the eastern U.S., which has elevated total Hg deposition compared to the broad central
region (see Figures 2-2 and 2-3).
As noted earlier, in order for a watershed to have elevated U.S. EGU-attributable risk, in the context of this risk
assessment, that watershed must have high U.S. EGU-attributable Hg deposition (as a fraction of total Hg
deposition) and elevated baseline fish tissue Hg concentrations.
53 Because the risk estimates for the 2005 scenario are being de-emphasized, we decided to only provide these
cumulative plots for the 2016 scenario. However, we note that the same relationship noted here (between U.S. EGU-
related Hg deposition for all -88,000 watersheds and for the subset included in the risk assessment), holds for the
2005 scenario.
79
-------
2.5 Overview of Risk Estimates
This section provides an overview of risk estimates generated for the 3,141 watersheds
included in the revised risk assessment. As noted earlier in Section 1.4.3, presentation of risk
estimates focuses on the typical female subsistence fish consumer scenario assessed at the
national-level, since this scenario provides the most comprehensive coverage for watersheds with
Hg fish tissue data across the U.S. and because the consumption rates used to model this scenario
represent subsistence levels supported by a number of studies. While this scenario is emphasized
in summarizing risk estimates, we also provide risk estimates for the other scenarios (i.e., low
income Blacks and Whites in the southeast, Tribal populations near the Great Lakes, Hispanics).
In summarizing risk estimates, we focus on the 2016 air quality scenario, since as discussed in
section 2.3, U.S. EGU-related Hg emission levels reflected in this scenario are closest to recent
conditions. The remainder of this section is organized as follows:
• Overview of percentile risk estimates generated for the fisher scenarios (Section 2.5.1):
We provide percentile risk estimates for the typical female subsistence fish consumer
scenario at the national level and summarize percentile risk estimates for the other SES-
differentiated female subsistence fish consumer scenarios assessed in the analysis. These
percentile risk estimates allow us to compare risk (total and U.S. EGU-attributable)
across the female subsistence fish consumers scenarios assessed in the risk assessment.
• Overview of the number (and frequency) of watersheds with populations potentially at-
risk due to U.S. EGU-attributable Hg deposition (section 2.5.2): This set of risk estimates
provides the main input to the risk characterization framework (see section 1.3).
Specifically, watersheds with populations potentially at-risk comprise:
o Watersheds where total risk is considered to represent a potential public health
concern and where U.S. EGUs contribute to that total risk. We considered various
increments of U.S. EGU contribution including > 5%, >10%, >15% and >20%,
although, we focus on cases where U.S. EGUs contribute >5 % as noted in section
1.3.
o Watersheds where risk from U.S EGUs alone (focusing on U.S. EGU deposition
and excluding other non-U.S. EGU sources) represents a potential public health
concern.
To support the discussion of risk estimates, we present a series of tables summarizing
those estimates and a list of observations based on this information.
2.5.1 Overview of percentile risk estimates (2016 scenario)
This section compares risk percentiles (for total and U.S. EGU-attributable risk) across
the set of female subsistence fish consumer scenarios included in the risk assessment, with an
emphasis on the typical female subsistence fish consumer scenario. We also compare and
contrast risk for the other female consumer scenarios with the more generalized typical female
subsistence fish consumer scenario.
80
-------
These percentile estimates are not population-weighted, but instead represent the specific
watershed-level risk estimates that fall at a specific point within the larger distribution of
watershed-level risk estimates for each female subsistence fish consumer scenario. In the
following tables, we sorted the total and U.S. EGU-attributable risk estimates separately for each
female subsistence fish consumer scenario. This means that the total and U.S. EGU-attributable
risk for a particular percentile are not matched by watershed (i.e., the 90th percentile total and
U.S. EGU-attributable risk for the Vietnamese subsistence fisher likely occur at different
watersheds).54 While this risk assessment focuses on subsistence levels offish consumption, the
risk tables also include risk estimates based on mean fish consumption rates for these female
subsistence fish consumer scenarios, which are relatively high compared to general recreational
angler rates but do not represent true subsistence rates.
Table 2-6 provides risk percentiles for RfD-based HQs for all of the female subsistence
fish consumer scenarios assessed in the analysis for the 2016 scenario.
Table 2-6 Percentile risk estimates for the full set of female subsistence fish consumer
scenarios included in the analysis (2016 scenario) (for both total and U.S. EGU
incremental RfD-based HQ)*
Fisher
consumption rate
(g/day) and
percentile* *
Watershed percentile risk (RfD-based HQ)*
Total
50th
75th
90th
95th
99th
U.S. EGU
50th
75th
90th
95th
99th
typical female subsistence fish consumer assessed nationally
39 (mean)
123 (90th)
173 (95th)
373 (99th)
1.9
6.1
8.5
18.4
3.5
10.9
15.3
33
5.6
17.7
24.9
53.7
7.4
23.4
32.9
70.9
11.2
35.4
49.7
107.2
-
0.2
0.2
0.5
0.1
0.3
0.4
0.9
0.2
0.5
0.7
1.5
0.2
0.7
1
2.1
0.4
1.2
1.7
3.7
Low income White female subsistence fish consumer in the Southeast
39 (mean)
93 (90th)
129 (95th)
286 (99th)
2.1
4.9
6.8
15.2
4.1
9.8
13.6
30.2
6.8
16.3
22.7
50.3
9.3
22.4
31
68.8
12.5
29.9
41.5
92
0.1
0.2
0.2
0.5
0.1
0.3
0.4
1
0.2
0.5
0.8
1.7
0.3
0.7
1
2.3
0.6
1.4
2
4.3
Low income Black female subsistence fish consumer in the Southeast
171 (mean)
446 (90th)
557 (95th)
(99th)
9.4
24.6
30.8
19.4
50.6
63.2
32.8
85.6
106.9
42.2
110.1
137.5
56.4
147.2
183.8
NC
0.3
0.8
0.9
0.6
1.6
1.9
1
2.7
3.4
1.4
3.6
4.4
2.6
6.8
8.4
NC
Low income Hispanic female subsistence fish consumer evaluated nationally
Presentation of non-matched percentile risk estimates represents a change from the March TSD. In that version,
we used risk bands to match trends in total risk to a specific U.S. EGU-attributable risk percentile. However, given
that total and U.S. EGU-attributable risk are not closely correlated, we determined that attempting to present
matched percentiles was of little utility. Instead, we have decided for this version of the Mercury Risk TSD to
simply provide percentile total and U.S. EGU-attributable risk estimates based on direct queries of the underlying
risk distributions.
81
-------
Fisher
consumption rate
(g/day) and
percentile* *
26 (mean)
98 (90th)
156 (95th)
(99th)
Watershed percentile risk (RfD-based HQ)*
Total
50th
1.2
4.5
7.2
75th
2.2
8.3
13.2
90th
3.5
13.3
21.2
95th
4.9
18.6
29.6
99th
8.3
31.5
50.1
NC
U.S. ECU
50th
0
0.1
0.2
75th
0.1
0.2
0.4
90th
0.1
0.4
0.7
95th
0.2
0.6
0.9
99th
0.3
1.1
1.7
NC
Vietnamese female subsistence fish consumer
27 (mean)
99 (90th)
152 (95th)
(99th)
1.1
4.1
6.3
2
7.2
11
3
11.1
17.1
4.1
15
23
6.8
24.9
38.3
NC
0
0.1
0.2
0
0.2
0.3
0.1
0.3
0.5
0.1
0.5
0.8
0.3
1
1.5
NC
Laotians female subsistence fish consumer
47 (mean)
145 (90th)
226 (95th)
(99th)
1.9
5.9
10.9
3.7
11.3
20.8
5.3
16.1
29.6
5.9
18
33
7.7
23.7
43.4
NC
0.1
0.2
0.3
0.1
0.3
0.5
0.2
0.6
1.2
0.3
0.9
1.7
0.9
2.6
4.9
NC
Tribal (near Great Lakes) female subsistence fish consumer
62 (mean)
136 (90th)
21 3 (95th)
493 (99th)
5.4
11.8
18.5
42.8
7.7
17
26.6
61.4
11
24.1
37.7
87.3
13.2
29.1
45.5
105.2
17.5
38.4
60.2
139.1
0.1
0.2
0.3
0.6
0.1
0.3
0.4
1
0.2
0.4
0.7
1.5
0.3
0.6
0.9
2.1
0.4
0.9
1.3
3.1
* Percentile risk estimates presented for "total" and "U.S. EGU" are not matched by watershed (e.g., the 90
percentile total and U.S. EGU-attributable risk estimates for a particular female subsistence fish consumer
scenario may are unlikely to be from the same watershed).
** Means are provided along with upper-end percentile values.
"-": RfD-based HQ is < 0.1.
NC: It was not possible to derive a 99thpercentile consumption rate for this population due to insufficient sample
size in the underlying study. Consequently, risk estimates for the 99th percentile consumption rates were not
generated.
We made several observations based on the information in Table 2-6 regarding the
potential health hazard associated with MeHg RfD-based HQ estimates (see Section 1.4.5).:
• Risk estimates for the typical female subsistence fish consumer (assessed at the national-
level) generally provide coverage for the Hispanic, Vietnamese and Tribal scenarios:
Risk estimates (for both total and U.S. EGU-attributable) generated for the typical female
subsistence fish consumer scenario are generally higher than estimates generated for the
Hispanic and Vietnamese scenarios. Risk estimates for the typical female subsistence fish
consumer scenario are also higher than risks for the Tribal population for U.S. EGU-
attributable risk although the Tribal populations tend to have higher total risks. Given
emphasis on U.S. EGU-attributable risk in the policy context, these trends would suggest
82
-------
that the typical female subsistence fish consumer scenario provides general coverage for
the other three scenarios (Hispanic, Vietnamese and Tribal), since the typical female
subsistence population tends to have higher U.S. EGU-attributable risk estimates.
• Risk estimates for the typical female subsistence fish consumer (assessed at the national-
level) may not provide full coverage for risks experienced by low income Blacks and low
income Whites in the Southeast, or for Laotians assessed at the national-level: U.S.
EGU-attributable risk estimates for the Southeastern low income White and low income
Black scenarios and for the Laotian scenario are higher than those for the typical female
subsistence fish consumer. In the case of low income Whites, this likely reflects the fact
that some regions in the Southeast (specifically South Carolina) have relatively elevated
total and U.S. EGU-attributable fish tissue Hg concentrations compared with the national
distribution (see Figures 2-13 and 2-15). In the case of low income Blacks, higher fish
tissue Hg concentrations in South Carolina are combined with fish consumption rates that
are substantially higher than those used for the typical female subsistence fish consumer
scenario (see Table 1-6). While both the low income White and low income Black
subsistence fisher scenarios in the Southeast have higher risk estimates than the typical
female subsistence fish consumer scenario (assessed nationally), because the typical
female subsistence scenario is assessed at all 3,141 watersheds with fish tissue Hg data,
we focus on the typical female subsistence fish consumer scenario to provided policy-
relevant risk information to support this rulemaking (i.e., we have greater overall
confidence in risk estimates generated for the typical female subsistence fish consumer
scenario because it is assessed at a larger number of watersheds with broader spatial
coverage). However, the higher risks estimates for these two Southeastern subsistence
fisher scenarios should also be considered in the context of interpreting the results of this
risk assessment in a science-policy context. In the case of the Laotian scenario, while
U.S. EGU-related risks are higher than for the typical female subsistence fish consumer
scenario, this scenario was assessed for a small number of watersheds (131 of the 3,141
watersheds with fish tissue Hg data). Specifically, risk estimates generated for the
Laotian scenario may reflect relatively rare localized interactions between fish tissue Hg
concentrations and U.S. EGU-related Hg deposition, resulting in areas of elevated U.S.
EGU-attributable exposure and risk.
2.5.2 Overview of number and percentage of watersheds with populations
potentially at-risk due to U.S. EGU mercury emissions (2016 scenario)
This section discusses risk estimates based on identifying the number and percent of
watersheds with fish tissue Hg samples with populations potentially at-risk due to Hg emitted
from U.S. EGUs (for the 2016 scenario). As noted in Section 1.3, the "at-risk population"
classification is based on identifying watersheds where: (a) U.S. EGUs contribute to total risk at
watersheds where that total risk is considered to represent a potential public health hazard and/or
(b) risk at the watershed-level represents a potential public health hazard from U.S. EGU-
attributable deposition when considered alone, without taking into account deposition from other
sources. The estimates of watersheds with at-risk populations discussed in this section are used
in the 2-Stage risk characterization framework described in Section 1.3 for interpreting risk
estimates. Specifically, the first category of at-risk populations described above comprise Stage
la of the 2-Stage approach, while the second category comprises Stage Ib (see Figure 1-1). The
83
-------
combination (i.e., mathematical union) of these two groups of watersheds with at-risk
populations comprises the set of watersheds represented in Stage 2 of the framework.
In summarizing risk estimates in this section, we focus on estimates generated using the
current risk assessment that reflects refinements to the risk model (e.g., inclusion of additional
fish tissue data and evaluating the typical female subsistence fish consumer for all watersheds
with fish tissue Hg data). However, when we present the Stage 2 risk estimates, we also include a
summary of Stage 2 estimates from the March version of the Mercury Risk TSD (then referred to
as "stage 3" results) for comparison.
With regard to the HQ estimates, any contribution of Hg from EGUs to watersheds with
potential exposures exceeding the MeHg RfD represents a hazard to public health, but for
purposes of this analysis, we have focused on those waterbodies where we determined EGUs
contributed 5% or more to the hazard. This is a reasonable, conservative approach because risks
are associated with all exposures above the MeHg RfD.
The watersheds with potentially at-risk populations are all based on the underlying risk
estimates generated for the typical female subsistence fish consumer scenario. The estimates of
watersheds with potentially at-risk populations are summarized in tables described below,
followed by a set of observations regarding these risk estimates.
• Tables 2-7: Identifies watersheds with potentially at-risk populations based on
consideration for different degrees of U.S. EGU contribution (i.e., >5, >10, >15 and
>20%) at watersheds where total risk is considered to represent a potential public health
hazard (i.e. HQ>1). For reference purposes, the table also identifies the total number of
watersheds (out of the 3,141 assessed for the typical female subsistence fish consumer)
with total risk exceeding the HQ threshold, regardless of the U.S. EGU percent
contribution (see the "> 0%" row of results in the table). In presenting results, the tables
include both the number and percent of watersheds with fish tissue Hg samples meeting
specific criteria that this represents. (Stage la of the 2-Stage framework)
• Table 2-8: Identifies watersheds with potentially at-risk populations based on U.S. EGU-
attributable deposition considered alone, without taking into account other sources of Hg
(Stage Ib of the 2-Stage framework)
• Table 2-9: Presents the union of the two categories of watersheds with potentially at-risk
populations (i.e., mathematical union of the Stage 1 and 2 estimates presented in Tables
2-7 and 2-8, see Figure 1-1 for explanation). Table 2-9 considers the number and percent
of watersheds that have (a) U.S. EGUs contributing to total risk of an HQ>1, OR (b) an
HQ>1 based on considering U.S. EGU Hg deposition alone, without taking into account
deposition from other sources (i.e., U.S. EGU-attributable HQ>1). This represents Stage
2 of the risk characterization framework.
• Table 2-10: Presents summary of the same Stage 2 risk metric as Table 2-9 with the
March Mercury Risk TSD results.
84
-------
Table 2-7 Watersheds with potentially at-risk populations based on consideration for
various degrees of U.S. EGU contribution to total risk (2016 scenario)
EGU risk
threshold
>0%
>5%
> 10%
> 15%
> 20%
Number and percentage of HUCs meeting risk threshold
criteria
(RfD-basedHQ>l)*
2016 analysis
90th percentile fish
consumption
2853
679
104
37
15
(91%)
(22%)
(3%)
-
-
95th percentile
fish
consumption
2983
726
114
40
17
(95%)
(23%)
(4%)
-
-
99th percentile fish
consumption
3112
754
119
40
17
(99%)
(24%)
(4%)
-
-
* Following convention for reporting HQ estimates to one significant digit, this requires an HQ > 1.5 (see
section 1.4.5). Results presented here for different levels offish consumption are each rank-ordered
separately (i.e., results are not matched for the same watershed across a given row).
Table 2-8 Watersheds with potentially at-risk populations based on consideration for risk
based on U.S. EGU mercury deposition and resulting exposure considered alone,
without taking into account other sources of mercury deposition (2016 scenario)
EGU risk
threshold
>1.5*
Number and percentage of 3,141 HUCs meeting risk
threshold criteria
(RfD-basedHQrisk)*
2016 analysis
90th fish
consumption
17
qcth percentile
fish
consumption
52 (2%)
(j(jth percentile jjsjj
consumption
327 (10%)
* Following convention for reporting HQ estimates to one significant digit, this requires an HQ > 1.5 (see
section 1.4.5). Results presented here for different levels offish consumption are each rank-ordered
separately (i.e., results are not matched for the same watershed across a the ">1.5" row).
Table 2-9 Combination of watersheds with potentially at-risk populations based on either
consideration for (a) U.S. EGU percent contribution to total risk OR (b) risk when
U.S. EGU mercury deposition is considered alone, without taking into account
deposition from other sources (2016 scenario)
EGU risk
threshold
Number and percentage of HUCs meeting risk threshold criteria
2016 analysis
90th percentile fish
consumption
95th percentile fish
consumption
99th percentile fish
consumption
U. S. EGU-attributable risk >1. 5*HQOR total risk >1.5*HQandU. S. EGU contribution of:
>5%
> 10%
> 15%
> 20%
679
107
47
29
(22%)
(3%)
-
-
744
141
80
64
(24%)
(4%)
(3%)
(2%)
917
385
339
331
(29%)
(12%)
(11%)
(11%)
*** Following convention for reporting HQ estimates to one significant digit, this requires an HQ > 1.5
(see section 1.4.5). Results presented here for different levels offish consumption are each rank-ordered
separately (i.e., results are not matched for the same watershed across a given row).
85
-------
Table 2-10 Reflecting the March version of the risk assessment - combination of
watersheds with potentially at-risk populations based on either consideration for (a)
U.S. EGU percent contribution to total risk OR (b) risk when U.S. EGU mercury
deposition is considered alone, without taking into account deposition from other
sources (2016 scenario)
EGU risk
threshold
Number and percentage of HUCs meeting risk threshold criteria
90th percentile fish
consumption
95th percentile fish
consumption
99th percentile fish
consumption
U. S. EGU-attributable risk >1. 5*HQOR total risk >1.5*HQandU. S. EGU contribution of:
>5%
> 10%
> 15%
> 20%
484
96
49
34
(20%)
(4%)
(2%)
-
524
121
76
61
(22%)
(5%)
(3%)
(3%)
672
325
292
286
(28%)
(14%)
(12%)
(12%)
* Following convention for reporting HQ estimates to one significant digit, this requires an HQ > 1.5 (see
section 1.4.5). Results presented here for different levels offish consumption are each rank-ordered
separately (i.e., results are not matched for the same watershed across a given row).
We made several observations regarding watersheds with potentially at-risk populations
due to U.S. EGU-attributable Hg deposition based on the information in Tables 2-7 through 2-
10, oriented around the 2-Stage risk characterization framework.
• Depending on the percentile fish consumption rate, between 3 and 24% of those
watersheds with total risk HQs >7 have U.S. EGUs contributing at least 5% of total Hg
deposition (2016 scenario): With this risk metric, we consider the degree to which U.S.
EGUs contribute to total risk at watersheds where total risk represents a potential public
health hazard (i.e., total exposure leads to an HQ > 1). Considering a 5% U.S. EGU
contribution at watersheds where total risk is considered a potential public health hazard,
we have up to 24% of the watersheds falling into this category (with the 24% value
reflecting risk modeled using the 99th percentile fish consumption rate for the high-end
female consumer - see Table 2-7). It is important when considering this risk metric to
reiterate that any exposure above the MeHg RfD represents a potential public health
hazard.
• Depending on the percentile fish consumption rate, between 2 and 12% of the watersheds
have HQs > 1, based on U.S. EGU Hg deposition before factoring in any other sources
ofHg (i.e., an U.S. EGU-attributable HQ>1) (2016 scenario): Our analysis suggests that
between 2 and 12% of the 3,141 watersheds modeled in the risk assessment for high-end
female consumers could have an HQ >1 from U.S. EGU-attributable Hg deposition when
considered alone, without taking into account other sources of deposition. The low end of
the range reflects the 95th percentile consumption rate, and the high end reflects the 99th
percentile consumption rate (again for the typical female subsistence fish consumer).
• Depending on the percentile fish consumption rate, between 22 and 29% of the
watersheds are at-risk based on at least one of the risk characterization criteria (2016
scenario): Combining the two categories of watersheds with populations at-risk due to
U.S. EGU Hg emissions summarized in the last two bullets, we get a total estimate
ranging from 22 to 29% of watersheds at-risk. These estimates are sensitive to the
86
-------
specification of the percent U.S. EGU contribution. If a 10 percent contribution threshold
is applied, the range of watersheds at-risk ranges from 3% to 12%. The ranges for each
percent contribution threshold also reflect the different fish consumption rates considered
for the high-end female consumer (i.e., 90th, 95th and 99th percentile fish consumption
rates). The results summarized here for total "at-risk" watersheds map to Stage 2 of the
2-stage risk characterization framework.
• The revised risk assessment and the March Mercury Risk TSD generate very similar
Stage 2 risk estimates (2016 scenario): Comparison of risk estimates presented in Tables
2-9 and 2-10 suggest that refinements to the risk assessment implemented since the
March TSD have not substantially affected overall Stage 2 results (formerly known as
"Stage 3" results in the March version of the Mercury Risk TSD).
2.6 Sensitivity Analyses
This section discusses several sensitivity analyses conducted to assess the potential
impact of sources of uncertainty related to: (a) the application of the proportionality assumption
in linking Hg deposition and fish tissue Hg concentrations (see sections 1.4.2.1 and 1.4.6) and (b)
use of the watershed-level 75th percentile fish tissue Hg concentration as the basis for generating
watershed-level risk estimates (see section 1.4.2). Given the emphasis placed in the national-
scale mercury risk assessment on risk estimates generated for the 2016 scenario, we have also
based the sensitivity analyses on simulating risks for the 2016 scenario.
To address uncertainty in applying the proportionality assumption, we conducted two
sensitivity analyses focused on concerns that the risk assessment may have included some
watersheds that are disproportionately impacted by non-air Hg sources. We included the
proportionality assumption in the sensitivity analysis because it represents a critical element of
the analysis and is acknowledged as representing a potentially important source of uncertainty
(see Table 2-15 in section 2.7).55 As noted in section 1.4.6, the proportionality assumption
linking Hg deposition over watersheds with fish tissue Hg concentrations, only holds for
watersheds where aerial Hg deposition is the primary source of loading. The two sensitivity
analyses include: (a) excluding four states where we have concerns over the potential for non-air
Hg playing a greater role (ME, MN, SC and LA)56 and (b) constraining the risk analysis to only
include those watersheds in the upper 25th percentile of total Hg deposition (i.e., watersheds with
55 In the March TSD, we included a sensitivity analysis on the proportionality assumption for flowing versus
stationary waterbodies. However, comments by the SAB suggested that this issue was unlikely to represent a
significant source of uncertainty in the analysis and consequently we have not repeated it here.
56 We excluded ME, MN and LA in this sensitivity analysis for reasons specific to each state. ME was excluded
because Hg fish tissue levels there are fairly high, while Hg deposition is not relatively elevated (compared to other
eastern states) - this raises the concern that other factor may affected fish tissue Hg (e.g., other non-air sources, or
increased methylation potential). MN was excluded because taconite mining could contribute non-air Hg loading.
Finally LA was excluded due to substantial industrial activity that could contribute non-air Hg loading. We note
that, in the March TSD, we excluded SC from this sensitivity analysis because of higher fish Hg levels and Hg air
deposition that (while elevated in some locations) is not uniformly higher than other states. As part of our
explanation for the higher fish tissue Hg concentrations in SC we pointed to a history of gold mining in SC (i.e., the
potential that significant non-air Hg loading may be a factor). SAB panel members with expertise in this area did
point out that the methylation potential of many of the waterbodies in SC (particular central and eastern SC) is
relatively high and that this is likely the reason for the higher fish tissue Hg concentrations.
87
-------
relatively elevated levels of total Hg deposition so we have increased confidence that aerial
deposition plays a significant role in loading).
To address uncertainty in use of the watershed-level 75th percentile fish tissue Hg
concentration in risk modeling, we conducted a sensitivity analysis using the watershed-level
median fish tissue Hg. This sensitivity analysis, which was recommended by SAB, considers the
situation in which subsistence fishers do not necessarily target the larger (higher trophic level)
fish and consequently could catch fish with lower fish tissue Hg concentrations.
The results of both sensitivity analyses are presented in terms of their impact on the 2-
Stage Risk Characterization Framework results (i.e., their impact on estimates of the number and
percent of'watersheds with potentially at-risk populations - section 1.3). Therefore, we provide
three tables (Tables 2-11, 2-12 and 2-13) that correspond to each stages of the 2-Stage Risk
Characterization Framework. Each table first presents core results, followed by results for each
of the three sensitivity analyses (this allows the results of each sensitivity analysis to be readily
compared against the core analysis). We present observations from the sensitivity analyses at the
end of this section. All results presented in the three tables are based on simulating risk for the
typical female subsistence fish consumer scenario.
Table 2-11 Sensitivity analysis results presented as: watersheds with potentially at-risk
populations based on U.S. EGVs making a specified contribution to total risk (2016
scenario)
ECU risk threshold
(percent U.S. ECU
contribution to risk)
Number and percentage of watersheds meeting risk threshold
criteria
(Toted HQ > 1.5 and U.S. EGU percent contribution as dimensioned
below)"
90th percentile fish
consumption
95th percentile fish
consumption
99th percentile fish
consumption
All watersheds (core analysis)
>0%
>5%
> 10%
> 15%
>20%
2853
679
104
37
15
(91%)
(22%)
(3%)
-
-
2983
726
114
40
17
(95%)
(23%)
(4%)
-
-
3112
754
119
40
17
(99%)
(24%)
(4%)
-
-
Sensitivity analysis A (exclude watersheds in MN, LA, SC and ME)
>0%
>5%
> 10%
> 15%
>20%
2104
638
90
32
14
(89%)
(27%)
(4%)
-
-
2224
685
100
35
16
(94%)
(29%)
(4%)
-
-
2347
713
105
35
16
(99%)
(30%)
(4%)
-
-
Sensitivity analysis B (include watersheds in top 25th percentile with regard to total Hg
deposition)
>0%
>5%
> 10%
712
154
39
(91%)
(20%)
(5%)
740
166
42
(94%)
(21%)
(5%)
776
178
45
(99%)
(23%)
(6%)
88
-------
ECU risk threshold
(percent U.S. ECU
contribution to risk)
> 15%
>20%
Number and percentage of watersheds meeting risk threshold
criteria
(Total HQ > 1.5 and U.S. EGU percent contribution as dimensioned
below)"
90th percentile fish
consumption
20
9
(3%)
-
95th percentile fish
consumption
20
9
(3%)
-
99th percentile fish
consumption
20
9
(3%)
-
Sensitivity analysis C (generate risks using median watershed-level fish tissue Hg
concentration)
>0%
>5%
> 10%
> 15%
>20%
2774
659
100
37
15
(88%)
(21%)
(3%)
-
-
2932
717
111
39
17
(93%)
(23%)
(4%)
-
-
3095
751
118
40
17
(99%)
(24%)
(4%)
-
-
* Following convention for reporting HQ estimates to one significant digit, this requires an HQ > 1.5 (see
section 1.4.5).
Table 2-12 Sensitivity analysis results presented as: watersheds with potentially at-risk
populations based on consideration for U.S. EGU-attributable HQ risk (2016 scenario)
(risk considering U.S. EGU Hg deposition before considering other sources ofHg
deposition)
U.S. EGU-attributable
risk threshold
Number and percentage of 3,141 watersheds meeting risk
threshold criteria
(U. S. EGU-attributable HQ) *
90th percentile fish
consumption
95th percentile
fish
consumption
99th percentile fish
consumption
All watersheds (core analysis)
>1.5*
17
-
52
Sensitivity analysis A (exclude watersheds in
>1.5*
7
-
22
(2%)
327
MN, LA, SC and ME)
-
209
Sensitivity analysis B (include watersheds in top 25th percentile with regard to
deposition)
>1.5*
9
-
25
(3%)
104
(10%)
(9%)
total Hg
(13%)
Sensitivity analysis C (generate risks using median watershed-level fish tissue Hg
concentration)
>1.5*
11
-
24
-
173
(6%)
* Following convention for reporting HQ estimates to one significant digit, this requires an HQ > 1.5 (see
section 1.4.5).
89
-------
Table 2-13 Sensitivity analysis results presented as: Combination of watersheds with
potentially at-risk populations based on either consideration for (a) U.S. EGU
percent contribution to total risk OR (b) risk when U.S. EGU mercury deposition is
considered alone, without taking into account deposition from other sources (2016
scenario)
EGU risk threshold
(percent U.S. EGU
contribution to risk)
Number and percentage of watersheds meeting risk threshold
criteria
(U. S. EGL '-attributable risk >1. 5*HQOR total risk >1.5*HQand
U.S. EGU: percent contribution as dimensioned below) *
90th percentile fish
consumption
95th percentile fish
consumption
99th percentile fish
consumption
All watersheds (core analysis)
>5%
> 10%
> 15%
>20%
679
107
47
29
(22%)
(3%)
-
-
744
141
80
64
(24%)
(4%)
(3%)
(2%)
917
385
339
331
(29%)
(12%)
(11%)
(11%)
Sensitivity analysis A (exclude watersheds in MN, LA, SC and ME)
>5%
> 10%
> 15%
> 20%
638
90
34
19
(27%)
(4%)
-
-
689
108
49
34
(29%)
(5%)
(2%)
-
782
267
221
213
(33%)
(11%)
(9%)
(9%)
Sensitivity analysis B (include watersheds in top 25th percentile with regard to total Hg
deposition)
>5%
> 10%
> 15%
> 20%
154
41
26
16
(20%)
(5%)
(3%)
(2%)
174
54
38
30
(22%)
(7%)
(5%)
(4%)
232
122
108
105
(30%)
(16%)
(14%)
(13%)
Sensitivity analysis C (generate risks using median watershed-level fish tissue Hg
concentration)
>5%
> 10%
> 15%
> 20%
659
101
42
23
(21%)
(3%)
-
-
721
118
53
36
(23%)
(4%)
(2%)
-
822
241
188
179
(26%)
(8%)
(6%)
(6%)
* Following convention for reporting HQ estimates to one significant digit, this requires an HQ > 1.5 (see
section 1.4.5).
Observations regarding the sensitivity analyses include:
Generating risk estimates excluding watersheds located in four states (AL, SC, ME and
MN): This sensitivity analysis resulted in different effects on the Stage la and Stage Ib
risk estimates, although neither was substantially different than the core estimates.
Estimates of the number of watersheds with at-risk populations due to U.S. EGUs
contributing at least 5% to total risk (Stage la) demonstrated a range of moderate
increases compared with the core analysis, depending on the "percent contribution"
category considered (see Table 2-11). Similarly, estimates of the percentage of
90
-------
watersheds with at-risk populations when considering U.S. EGU Hg deposition (without
other sources - Stage Ib) based on dropping out the four states was 9% as compared with
10% when considering all states (for the 99th percentile fish consumption rate - see Table
2-12). This sensitivity analysis suggests that the risk estimates are relatively robust to
exclusion of the four states considered in this sensitivity analysis.
• Generating risk estimates including only those watersheds falling in the top 25th
percentile with regard to total Hg deposition: These sensitivity analysis results suggest
that focusing on those watersheds with higher total deposition results in a mixture of
impacts on both Stage la and Stage Ib risk metrics. Estimates of the number of
watersheds with at-risk populations due to U.S. EGUs contributing between 5% and 10%
is lower for the sensitivity analysis than for the core analysis. However, estimates for
contributions of > 10% and >20% are notably higher than for the core analysis (see Table
2-11). Similarly, estimates of the percentage of watersheds with at-risk populations when
considering U.S. EGU Hg deposition alone (without taking into account deposition from
other sources - Stage Ib) is 13% as compared with 10% when considering all watersheds
(for the 99th percentile fish consumption rate - see Table 2-11). This sensitivity analysis
suggests that the risk estimates would be somewhat higher (particularly for the tail of the
risk distribution) if we focus on those watersheds receiving the greatest amount of
atmospherically deposited Hg. However, the magnitude of risk is not substantially
different from the core analysis.
• Generating risk estimates using the median watershed-level fish tissue Hg concentration
(instead of the 75th percentile as used in the core analysis): These sensitivity analysis
results are notably different for the two categories of risk metric considered. Estimates of
the number of watersheds with at-risk populations due to U.S. EGUs contributing > 5%
are essentially the same as the core risk estimates. However, estimates of the percentage
of watersheds with at-risk populations when considering U.S. EGU Hg deposition
(without other sources - Stage Ib) is substantially lower than the core analysis (i.e., 5%
when the median fish tissue Hg concentration is used as compared with 10% with the
core analysis). These results suggest that for some risk metrics, the results of the analysis
would be significantly affected were we to use the median fish tissue Hg concentration
rather than the 75th percentile. The SAB peer-review panel concluded that "using the 75th
percentile offish tissue values as a reflection of consumption of larger, but not the
largest, fish among sport and subsistence fishers is a reasonable approach and is
consistent with published and unpublished data on predominant types offish consumed."
2.7 Discussion of key sources of variability and uncertainty
This section provides a qualitative discussion of variability and uncertainty associated
with the risk assessment. Regarding variability, we focus on identifying the key sources of
variability associated with modeling risk for the scenarios included in the analysis and then
discuss the degree to which those sources are reflected in the design of the risk assessment and
implications for the risk estimates generated. The risk assessment has been designed to reflect
critical sources of variability to the extent allowed by available methods and data and given the
resources and time available. Key sources of variability associated with the analysis include:
91
-------
• The pattern of total and U.S. EGU-attributable Hg deposition across watersheds in the
U.S. (including variation in the global inflow of Hg into the U.S.)
• The patterns of fish tissue Hg concentrations across watersheds within the U. S.
• The sampling protocols for measuring fish tissue Hg concentrations used by States which
results in varying degrees of spatial coverage offish tissue Hg concentrations across the
states.
• The response of watersheds to Hg loading including variation in the temporal response
and degree to which fish tissue Hg concentrations change in response to altered Hg
loadings.
• The degree of subsistence fishing activity by different groups across watersheds.
• Subsistence fishing behavior including: consumption rates, species harvested, degree to
which activity is focused on one or more waterbodies and differences in
cooking/preparation practices. Variation in bodyweight can also impact exposure and risk
estimates.
These sources of variability and the degree to which each is reflected in the design of the
analysis are identified and described in Table 2-14.
Regarding uncertainty, we focus first on identifying potentially significant sources of
uncertainty impacting the analysis. Then we characterize (a) the nature of the impact of each
source on risk estimates and (b) the degree to which the potential impact of the source of
uncertainty is characterized as part of the analysis (including whether sensitivity analyses
completed for the risk assessment address a particular source of uncertainty). The results of this
qualitative assessment are presented in Table 2-15.
The SAB commented extensively on the sources of variability and uncertainty that should
be addressed in the revised version of this section and those recommendations are reflected in
Tables 2-14 and 2-15.
92
-------
Table 2-14 Key sources of variability associated with the analysis and degree to which they are reflected in the design of the
analysis
Source of variability
Description
Degree to which source is reflected in design of risk analysis and implications
for the risk estimates generated
(A) 2005 Hg emissions
Emissions approaches have been applied to create a 2005
Hg inventory for separate categories of emissions (e.g.,
Coal utilities, Portland Cement, Electric Arc Furnaces).
Within each category, types of variability for total Hg
include (1) annual variability related to product demand,
(2) sub-annual (e.g., hourly) temporal variability from
production related fluctuations at each facility, (3)
variability of emissions control systems over time
depending on the control device age and process stages
such as startup/shutdown and operating conditions even
during standard operations, (4) fuel characteristics
including the Hg content of the fuel, and (5) differences in
similar pieces of equipment that are relevant when
emission factors are applied to a class of process (e.g., not
all fluidized bed boilers will have the same emissions
potential). Further, the speciation characteristics of the
total Hg has variability based on many of the same factors
as for total Hg (e.g., production parameters, control
systems, fuels, physical properties of individual units)
The analysis accounts for annual variability in that estimates of Hg emissions were
developed for the particular 2005 year as the midpoint of the fish tissue data
projected to 2016 to estimate risks in the future baseline. The analysis attempted to
estimate sub-annual variability through use of (a) hourly EGU demand data that is
developed from several years of hourly Continuous Emissions Monitoring (CEM)
heat input data and (b) monthly, day-of-week, and diurnal temporal allocation factor
for non-EGU sources. The approach for estimating 2005 EGU emissions was tied to
1999 control assumptions and fuels, which creates some uncertainty described in
Table 2-15. The variability associated with different pieces of equipment in a class
of process was not accounted for explicitly, but the Hg emission factors developed
for EGU boilers and other unit types were developed to represent, on average,
emissions from a given unit type by averaging emissions test data over a many units
at facilities within a given class. The emission factors used to compute Hg for U.S.
EGUs were assigned based on boiler type, fuel (coal or oil), and control device
configuration. The same characteristics were used to assign speciation fractions.
(B) 2016 Hg emissions
The sources of variability identified for the 2005
emissions also apply to 2016; however, they are addressed
differently because of the differences in approach
associated with estimating emissions for a future year.
Annual variability is accounted for by the Integrated Planning Model's 2016 estimate
of Hg emissions. This model accounts for expected electricity demand, controls in
place before 2016, and forecasts of fuel prices resulting in predicted choices of fuels.
Further, the model predicts the Hg content of the fuel used in each unit in the future
based on forecasts of which coal mine will supply coal to each unit. Thus, IPM does
a more complete job than is done in 2005 for including control and fuel information.
As in 2005, sub-annual variability uses a) hourly EGU demand data that is developed
from several years of hourly Continuous Emissions Monitoring (CEM) heat input
data and (b) monthly, day-of-week, and diurnal temporal allocation factor for non-
EGU sources. IPM further accounts for differences in equipment using Emissions
Modification Factors (EMFs), which adjust the Hg emissions based on the boiler
type and control device configuration. The EMFs are not specific to each unit, but
rather were developed to represent, on average, emissions changes due to boiler and
control device types. Finally, the speciation fractions used in 2005 were also applied
in 2016, but the fractions were assigned based on the 2016 unit type, fuel type, and
control devices, rather than the base year configuration.
(C) Pattern of total and
U.S. EGU-attributable
Hg deposition across
Patterns of annual deposition of Hg including total (all
source) and estimates of the U.S. EGU fraction (by
watershed) displays considerable spatial variability across
By extrapolating CMAQ grid cell results at the more spatially refined HUC12
watershed level, we retain the greater degree of spatial resolution in characterizing
Hg deposition obtained through the use of the 12k CMAQ grid cell simulations.
93
-------
Source of variability
Description
Degree to which source is reflected in design of risk analysis and implications
for the risk estimates generated
watersheds in the US
(including variation in
the global inflow of Hg
into the U.S.)
the U.S. based on the results of 12 km grid cell CMAQ
modeling (see Section 2.4). This variability includes
changes over time in global-scale ambient mercury
concentrations due to changes in global emissions and
chemistry.
Regarding variation in the global inflow of Hg, the global GEOS-CHEM simulation
does provide ambient Hg inflow estimates used in the CMAQ model domain that
vary spatially in the horizontal direction and vertically from the surface through the
troposphere. The Hg inflow concentrations also vary for each day of the year and
within each day. However, the ambient speciated Hg inflow is kept constant between
2005 and 2016 scenarios.
(D) Patterns of fish
tissue Hg
concentrations across
watersheds within the
U.S.
Mercury fish tissue measurements can display
considerable spatial variability across watersheds (see
section 2.3).
We have fish tissue measurements for roughly 4% of the watersheds in the U.S. (i.e.,
3,141 watersheds with measured values out of 88,000 based on data from 2000-
2010). These measurements are concentrated in the Eastern part of the U.S., although
there are some measurements in almost all states. While our measured fish tissue
levels do generally provide some degree of coverage for areas with elevated Hg
deposition (in terms of both total and U.S. EGU-attributable), this coverage is limited
and there are a large number of watersheds with high total and U.S. EGU-attributable
Hg deposition for which we do not have fish tissue measurements (see section 2.4).
(E) Sampling protocols
for measuring fish
tissue Hg
concentrations used by
States differ, which
results in varying
degrees of spatial
coverage offish tissue
Hg concentrations
across the states.
States utilize varied approaches for targeting waterbodies
for fish sampling and Hg measurement. In virtually all
cases, sampling strategies are not intended to provide a
representative characterization offish tissue Hg
concentrations across the state and instead are intended to
identify and then target areas suspected of having higher
Hg levels, or higher fishing activity.
The sampling protocol used by states in collecting fish tissue Hg measurements is
directly reflected in the fish tissue dataset used in the risk assessment. Therefore, any
potential bias in the collection offish tissue data (i.e., tendency to capture more
highly impacted waterbodies) is reflected in that dataset. However, having fish tissue
Hg data that are potentially biased towards more heavily Hg-impacted waterbodies is
not problematic given the focus of the risk assessment. Specifically, given that we are
attempting to capture reasonable high-end subsistence fisher risk, having fish tissue
data biased towards more highly impacted waterbodies is actually preferable, since it
increase the potential that we will identify areas of higher exposure and risk related
to the subsistence fisher scenario (it would only be problematic if we were attempting
to generate a representative picture of more generalized recreational fisher risk).
(F) Response of
watersheds to Hg
loading including
variation in the
temporal response and
degree to which fish
tissue Hg
concentrations change
in response to altered
Hg loadings.
The response offish tissue Hg concentrations to mercury
deposition can vary greatly depending on a number of
factors (e.g., role of watershed in loading to waterbody,
methylation potential of the waterbody, rates of sulfate
deposition, nature of aquatic biotic foodweb including mix
of upper-level trophic fish etc). These factors can also
affect the temporal profile of that response.
Variation in methylation potential is reflected directly in the variation in the
measured Hg levels in fish across the different sampling locations. This variation is
very important, and in fact is a primary determinant of variation in fish tissue Hg
concentrations across watersheds. This will have a large impact on the variability in
total Hg exposure, and as shown in Figure 2-17, results in a low correlation between
total Hg deposition and fish tissue Hg across watersheds. However, this variability
will not have an impact on the attribution of exposure at any specific watershed,
because the scientific literature supports a linear relationship between changes in
deposition of Hg at a watershed and changes in fish tissue Hg at the same watershed.
Furthermore, because we are not predicting temporal trends in fish tissue levels and
instead consider a future point in time (once near-steady state conditions are
reached), variation in the temporal profile of changes in fish tissue levels related to
differences in methylation potential of different watersheds is also minimized as a
factor in the analysis.
(G) Differences in the
spatial distribution and
Studies reviewed in developing the approach for this
analysis suggests that there can be considerable variation
Surveys of high- self-caught fish consuming populations allow us to clearly define
this type of activity for specific areas covered by those surveys (e.g., Hispanic,
94
-------
Source of variability
Description
Degree to which source is reflected in design of risk analysis and implications
for the risk estimates generated
extent of subsistence
fishing activity by
different groups across
watersheds in the U.S.
in the nature of high-end self-caught fisher populations
across regions of the country. This variation reflects ethnic
and cultural practices and can also be driven be socio-
economic status (SES) related factors (e.g., poverty, levels
of income). In addition, we would expect that access to
freshwater vs saltwater fishing locations would also play a
role.
(H) Subsistence fishing
behavior including:
consumption rates,
species harvested,
degree to which
activity is focused on
one or more
waterbodies and
differences in
cooking/preparation
practices. Variation in
bodyweight can also
impact exposure and
risk estimates.
Vietnamese and Laotian fishing activity in specific regions of California, high-end
fishing activity by Blacks and Whites in South Carolina and Tribal activity near the
Great Lakes). However, available studies on this type of high-end fishing activity at
inland freshwater bodies do not provide comprehensive coverage for all regions in
the U.S. Therefore, we extrapolated coverage from areas reflected in these studies to
other portions of the U.S. There is uncertainty associated with this process. As
discussed in Table 2-15 (entry E) a lack of comprehensive survey information on
subsistence fishing activity for all regions in the U.S. is a limitation of the analysis.
This also prevents us from generating population-weighted estimates of risk.
Subsistence fisher populations can display significant
variability in their fishing behavior, which can impact
overall exposure and risk. Variation in consumption rates
has a direct impact on exposure, as do differences in the
types of fish (size and species) potentially targeted.
Tendency to focus fishing activity in one spot, versus
distributing it across different waterbodies can also impact
risk. Specifically, activity distributed across multiple
waterbodies will tend to dilute out locations with higher
fish tissue Hg concentrations, while focused activity
means that one or more fishers may get all of their fish
from the more impacted lake. Differences in
cooking/preparation can also impact Hg levels on a per
unit of fish basis, which, when combined with "as cooked"
consumption rates can effected modeled exposure and
risk. Body weight will also affect the calculation of
exposure, when standardized by body weight (e.g., a
heavier individual ingesting the same amount offish will
have a lower body-weighted adjusted exposure than a
lighter individual ingesting the same amount offish).
While we have fairly comprehensive information on consumption rates for the set of
female subsistence fish consumer scenarios we included in the analysis, we do not
have information (for those fishers) characterizing several of the other behavior-
related factors including types of species and size of fish targeted and the degree to
which fishers target a single waterbody, versus distribute their activity across
multiple waterbodies. However, given the focus of this assessment on capturing
reasonable high-end fish consumption by subsistence fishers, we have modeled these
scenarios assuming that (a) they tend to favor somewhat larger fish (use of the 75th
percentile watershed-level fish tissue Hg concentration) and (b) they focus their
fishing activity on a single watershed. While we do not know how representative
these types of higher-exposure related behaviors are within the subsistence fisher
populations covered in the risk assessment, we do believe it is reasonable to assume
that some subset of each population could engage in this type of behavior. With
regard to fish cooking/preparation, we have not considered variability in this
behavior, although the factors used (1.5- see section 1.4.4), is a value that reflects
the range of adjustment factors seen with different types of fish preparation methods
for different types of fish. Similarly for body weight, we have used a central
tendency factor and not explicitly modeled variability in this factor.
95
-------
Table 2-15 Key sources of uncertainty associated with the analysis, the nature of their potential impact on risk estimates, and
degree to which they are characterized as part of the analysis
Source of
uncertainty
Description
Nature of potential impact on the exposure and risk
estimates
Degree to which the potential impact of the source of
uncertainty is characterized as part of the analysis
Characterizing the pattern offish tissue Hg concentrations across watersheds in the U.S. (reflects small fraction of total watersheds with fish tissue measurement data,
differences in state-level protocols for collection offish tissue data, substantial number of watersheds with relatively low sample sizes (e.g., 1-2 samples), uncertainty associated with
filtering watersheds to exclude locations with potentially significant non-air Hg impacts)
(A) Relatively small
fraction of U.S.
watersheds with
measured fish tissue
He data
Fish tissue Hg measurements are
available for 3,141 of the -88,000
watersheds in the U.S. (i.e., -4%
of the watersheds). This relatively
low coverage raises concerns that
the risk assessment may miss
areas of elevated U.S. EGU Hg
deposition and/or areas with
elevated fish tissue Hg
concentrations, both of which
could result in a low-biased
assessment of U.S. EGU-related
risk.
Supplemental data for Pennsylvania that has been
incorporated into the fish tissue dataset to enhance the
current version of the risk assessment reduces
somewhat concerns over missing areas of high U.S.
EGU deposition (Pennsylvania has high U.S. EGU
deposition and in the March version of the risk
assessment had very little coverage by fish tissue Hg
data). Similarly, additional data for Wisconsin and
Minnesota reduces concerns over missing areas with
elevated fish tissue Hg concentrations. However, there
is still the potential that areas of high U.S. EGU Hg
deposition and/or areas with high fish tissue Hg
concentrations have been excluded from the analysis. In
addition, there is still an overall low bias in our
estimates of the number of watersheds with specified
levels of risk since only a fraction of watersheds have
fish tissue Hg measurements.
We have send-quantitatively examined the issue of
potentially missing Hg deposition hot spots by presenting
Figures 2-15 and 2-16, which show coverage by fish tissue
Hg data for areas of the country with elevated U.S. EGU
Hg deposition. Examination of these maps suggests, for
example, that while Pennsylvania is now fairly well
covered by available fish tissue data, Illinois, Kentucky
and Ohio are less well covered. It is also important to point
out that even while general coverage for high U.S. EGU
Hg deposition areas by measured fish tissue data has
improved for this version of the risk assessment, the
potential still exists that substantial numbers of high-
impact watersheds have not been included in the risk
assessment due to a lack of fish tissue sampling at those
locations.
(B) Differences in
state-level protocols
for collection of fish
tissue data which
results in differing
degrees of spatial
coverage across states
and differing degrees
of potential bias in the
samples collected
States use different strategies for
determining where fish tissue
samples will be taken
geographically as well as the
types of fish that will be measured
(and the methods of collection).
Generally, however, there is likely
to be a bias towards targeting
waterbodies suspected of having
elevated fish tissue Hg
concentrations for most states.
Differences in the strategies used
to collect fish tissue data together
with the potential for bias
favoring more highly impacted
waterbodies, introduces
uncertainty into the risk
assessment. Specifically, there is
Because the goal of the risk assessment is to capture
risk for subsistence fishers that are likely to experience
elevated U.S. EGU-attributable risk, having fish tissue
Hg concentrations that likely reflect targeting of more
highly impacted waterbodies (i.e., biased towards
higher risk locations) is not problematic and in fact, is
preferable. If it is reasonable to assume that subsistence
fisher activity could occur at these sampled watersheds
that have relatively higher fish tissue Hg concentrations,
then having risk estimates based on these measurements
strengthens the analysis from a science policy
standpoint. By contrast, if the analysis had been
designed to generate a fully representative population-
weighted assessment of subsistence fisher risk, then
bias in the fish tissue Hg concentrations would be
problematic (since it would high-bias an assessment
targeted at capturing "typical" risk for these high-
consumers)^
Given that potential high-bias in the characterization of
fish tissue Hg concentrations across the watersheds
included in this risk assessment is not problematic and
may in fact be preferable, we did not attempt to quantify
the impact of this factor on risk estimates.
96
-------
Source of
uncertainty
Description
Nature of potential impact on the exposure and risk
estimates
Degree to which the potential impact of the source of
uncertainty is characterized as part of the analysis
likely to be varying degrees of
high-bias in the characterization
of fish tissue Hg concentrations
across states.
(C) Substantial
number of watersheds
with relatively low
sample sizes (e.g., 1-2
samples)
Only 28% of the watersheds with
fish tissue Hg data have at least
10 samples (-41% of the
watersheds have only 1 or 2
samples). Generating 75th
percentile risk estimates for
watersheds with a small number
of samples (e.g., <10 to 20) is
subject to uncertainty and will on
average understate the true 75th
percentiles across watersheds,
although the bias in the estimate
of the 75th percentile for any
specific watershed is unknown.
The potential that a substantial fraction of the 3,141
watersheds included in the risk assessment could have
low bias in the 75th percentile fish tissue Hg
concentrations used in modeling risk means that in turn,
risk estimates for a substantial number of watersheds
could be low-biased. For the very low samples sizes
(e.g., 1-2 samples per watershed), it is likely that, as a
trend, the estimates of the 75th percentile values are
actually more likely representative of the central
tendency, e.g. the median.
We examined trends in the watershed-level 75 percentile
estimates across different strata of watersheds, where those
strata were based on sample size (see Table 1-2). As
discussed in section 1.4.2, the average of the 75th
percentile estimates increases substantially across these
strata (i.e., as the number of samples available to
calculated a 75th percentile increases, the trend in 75th
percentiles also increases). This suggests that other factors
equal (see below), there is a distinct potential for low bias
in the 75th percentile fish tissue Hg levels used in risk
characterization for those watersheds with lower sample
sizes. The important caveat to this is that this observation
assumes that there is no correlation between absolute level
offish tissue Hg concentrations and sample size across
watersheds. This may not be the case if, for example,
waterbodies with higher absolute fish tissue Hg
concentrations are also targeted more for sampling (which
is likely the case). Therefore, the potential for low-bias in
the 75 percentile estimates for watersheds with 1-2
samples probably needs to be softened slightly since there
is the potential for these watersheds to have lower "actual"
measured fish tissue Hg levels.
(D) Filtering
watersheds to exclude
locations with
potentially significant
non-air Hg impacts
As described in section 1.4.2, we
filtered watersheds (with fish
tissue Hg data) to exclude those
potentially impacted by non-air
Hg sources including gold mines
and any industrial sources
meeting a specified Hg release
threshold. However, we did not
filter out watersheds located near
large urban areas which have the
potential to release Hg through
waste water treatments effluent.
Recall that the proportionality
assumption used as the basis for
generating U.S. EGU-relevant
exposure and risk does not hold in
In those instances where risk is modeled for a watershed
assuming that air Hg deposition is dominant, when in
reality, municipal wastewater discharge may be
contributing significantly to Hg loading, then estimates
of U.S. EGU-attributable risk for that watershed could
be biased high. In this case, we would not have assigned
a portion of risk to the municipal discharges and instead
would have over-estimated the U.S. EGU-attributable
fraction of risk.
Although we did not explicitly consider the issue of urban
wastewater Hg loading in our sensitivity analyses, two of
the sensitivity analyses did examine the broader issue of
uncertainty in filtering our watersheds to exclude areas
with substantial non-air Hg loading (se section 2.6). These
two sensitivity analyses suggested that the risk estimates
generated in the core analysis are relatively robust even
when we (a) exclude a set of states that may have
substantial non-air Hg loadings across their watersheds (or
have higher methylation potentials in their waterbodies) or
(b) focus risk modeling on those watershed in the upper
quartile with regard total Hg loading (and therefore are
more likely to be the dominant source of Hg to a
watershed).
97
-------
Source of
uncertainty
Description
Nature of potential impact on the exposure and risk
estimates
Degree to which the potential impact of the source of
uncertainty is characterized as part of the analysis
instances where non-air Hg
loading is substantial.
Characterizing subsistence fishing activity across areas of high U.S. EGU mercury deposition (includes assessing potential locations for subsistence fishing activity as well as
details regarding actual fishing and fish consumption behavior)
(E) Predicting which
watersheds are likely
to experience
subsistence fishing
activity
There is uncertainty associated
with estimating which of the
watersheds for which we have
fish tissue Hg data are likely to
experience subsistence fishing
activity associated with the female
subsistence fish consumer
scenarios considered in the
analysis. For the typical female
subsistence fish consumer
scenario, we assumed that activity
could occur at any of the
watersheds with fish tissue data.
However, for the remainder of the
female subsistence fish consumer
scenarios, we used the concept of
"source population" (see section
1.4.3) to guide identification of
watersheds with the potential for
subsistence fisher activity by
these specific SES-differentiated
groups of fishers.
Because the goal of the analysis is to determine whether
there is the potential for significant risk for female
subsistence fish consumers and not to generate
representative population-weighted risk distributions for
those populations, the importance of rigorously
assessing where these populations are active (and the
numbers of fishers at each location) is reduced
somewhat. However, bias could be introduced into the
analysis if we have modeled risk for high-impact
watersheds when in reality there is no likelihood of
subsistence fishing activity at those locations. The
potential impact of this source of uncertainty is difficult
to assess. However, as discussed above, because we are
not attempting to generate population-weighted risk
distributions, it is not expected to substantially impact
the risk assessment.
We did not explicitly examine this source of uncertainty
related to where specific subsistence fisher populations
may be active.
(F) Focus on risks to
subsistence fishing
populations and
associated female
subsistence fish
consumers (concern
that analysis does not
cover high-end
recreational angler
risk and risk by other
potentially high-
consuming groups)
The SAB raised concerns that by
focusing on subsistence fishers,
we may have not provided
coverage in the risk assessment
for other high-consuming
populations including the subset
of recreational anglers who
frequently fish and consume the
fish they catch, and thus have near
subsistence level consumption
rates.
The typical female subsistence fish consumer scenario
is actually characterized (in terms of consumption rates)
using a survey of recreational anglers conducted in
South Carolina (Burger et al, 2002). This population is
characterized as "subsistence" for purposes of this
analysis due to the magnitude of the upper percentile
consumption rates used (i.e., the 90th to 99th percentile
values) which are considered subsistence since they
represent a substantial contribution to dietary intake of
protein. Furthermore, given that in this revised analysis,
the typical female subsistence fish consumer scenario
was assessed at all watersheds with fish tissue Hg data
(excluding those filtered due to high-non air Hg
impacts), this scenario would provide complete
coverage for high-end recreational anglers, both in
terms of consumption rate and potential location of
We did not examine this issue of providing coverage for
high-end recreational anglers because, as discussed in the
cell to the left, the typical female subsistence fish
consumer scenario provides coverage for a high-end
recreational angler in the context of this assessment (i.e.,
given the goal of assessing subsistence-level fish
consumption at each watershed without population
weighting).
98
-------
Source of
uncertainty
Description
Nature of potential impact on the exposure and risk
estimates
Degree to which the potential impact of the source of
uncertainty is characterized as part of the analysis
activity. There is no systematic bias introduced into the
analysis due to its focus on female subsistence fish
consumer scenarios that is consistent with the overall
goal of the analysis (i.e., to characterize risk for the
most highly impacted fisher populations).
(G) Characterizing
subsistence fisher
behavior (including
consumption rates,
types/size of fish
targeted and the
degree to which
activity is focused at
specific waterbodies)
There is uncertainty associated
with characterizing subsistence
fisher behavior including
consumption rates, types and size
of fish targeted and the degree to
which fishers target specific
watersheds rather than distribute
their activity across multiple
watersheds.
There are relatively comprehensive survey data
available for characterizing fish consumption rates for
the subsistence scenarios included in the analysis. In
fact, the subsistence fisher scenarios included in the
analysis were selected, in large part, based on the
availability of survey data to characterize consumption
rates. In addition, the consumption rates display a
reasonable degree of consistency in terms of magnitude
across the studies (see Table 1-6). Therefore, we believe
that there is relatively low uncertainty associated with
characterizing this aspect of subsistence fisher behavior
and little concern for the introduction of bias into the
analysis.57
However, the other critical aspects (degree to which
fishers target specific sizes and species of fish and the
degree to which their activity is focused at one or more
watersheds) are not well characterized in the literature
and therefore are subject to considerable uncertainty.
However, because the analysis is focused on modeling
reasonable high-end risk, the importance of uncertainty
associated with these additional behavior-related factors
is reduced substantially. Specifically, if it is reasonable
to assume that a subset of subsistence fishers would
Given that the focus of this analysis is on assessing risk for
populations likely to experience the highest reasonable
U.S. EGU-attributable risk, we concluded that because it is
reasonable to assume that some fraction of high-end
fishing populations could focus their activity on a single
watershed (and could favor larger fish), we can model this
behavior in our analysis without introducing substantial
uncertainty. If these assumptions regarding behavior are
relaxed (i.e., fishers were assumed to eat smaller fish and
distribute their activity across multiple watersheds), then
risk will be reduced.* However, we did not explicitly
model these alternative behavioral profiles given our focus
on capturing reasonable estimates of high-end risk.
* The observation that distributing fishing activity between
watersheds will reduce risk needs additional clarification.
While distributing fishing activity between watersheds
may or may not impact central tendency estimates of risk
across a group of fishers, high-end risk (risk for
individuals at the upper tail of a simulated distribution)
will be lower, since instances of having an individual fish
at a single high Hg watershed will be removed, with
fishing activity at that high Hg watershed now being
averaged with activity at other less-impacted watersheds.
Concerns have been raised by the SAB that some of the studies providing the consumption rates are older and that therefore, these studies may not capture current fishing
practices. In addition, concerns were raised that sampling frequencies characterizing higher consumption levels for some of the fishing populations were relatively low (e.g., Black
fishers in the Burger et al., 2002 study). The sources of uncertainty described here would have a potentially significant impact on the risk assessment if we were attempting to
generate a representative picture of the actual distribution of risk across the full body of fishers reflected in each scenario. In that case, errors related to the study being old, or
having insufficient survey samples to full characterize specific high-end percentiles of consumption would be a significant concern. However, when the goal of the analysis is to
identify a set of high-end subsistence-like fish consumption rates that we believe could exist for a subset of each fishing population (and not to generate a representative
population-weighted picture of risk), then the potential impact of these sources of uncertainty is reduced. Concerns were also raised by the SAB that the fish consumption surveys
could reflect seasonally-differentiated consumption rates (i.e., higher rates may only occurred for part of the year during increased fishing activity). Regarding this issue of
seasonality in the consumption rates, our evaluation of all three studies suggests that they provide annualized consumption rates (see Table 1-6) and therefore, we believe that the
likely impact of this source of uncertainty is limited.
99
-------
Source of
uncertainty
(H) Estimating unit
concentration of
MeHg in cooked fish
(application of
adjustment factors for
the fraction of total
Hg that is MeHg in
fish and cooking/
preparation).
Description
Two adjustment factors are
applied to the 75th percentile fish
tissue Hg concentration at a given
watershed to generate an estimate
of MeHg in the cooked fish
serving (in ppm). These include a
Hg conversion factor for
estimating the fraction of Hg in
fish that is MeHg and a fish
preparation/cooking adjustment
factor for adjusting the MeHg
concentration in the fish to reflect
preparation of the fish. There is
uncertainty associated with both
of these factors.
Nature of potential impact on the exposure and risk
estimates
target larger fish and focus their fishing activity at a
specific watershed, then uncertainty in these two factors
is largely ameliorated.
Mercury conversion factor: the factor used in this
analysis (0.95) reflects consideration for the range of
values provided in the Mercury Study Report to
Congress (U.S. EPA, 1997) (i.e., >0 .90) (see section
1.4.4). It would be possible to include a sensitivity
analysis considering 0.90 and 0.99, but given that this
factor is linearly related to both exposure and risk, the
outcome would be similarly modest: approximately 5%
lower and higher risk, respectively for a given
watershed- specific estimate. Given available
information, we do not believe that any systematic bias
is introduced into the analysis due to the use of this
parameter value.
Fish preparation/cooking adjustment factor: The factor
used in this analysis (1.5) reflects consideration for the
range of preparation factors provided in Morgan et al.,
1997) (i.e., 1.1 to 1.5 for walleye and 1.5 to 2.0 for lake
trout - see section 1.4.4). Again, as with the Hg
conversion factor, we could consider a range of factors
as a sensitivity analysis, but again, given that exposure
and risk are linearly related to this factor, the outcome
would is predictable. In this case, the impact on risk
could be more pronounced, since the potential factors
range from 1.0 to 2.0. The SAB identified several
alternative studies for characterizing this factor,
suggesting that those studies would support a smaller
(or no) positive adjustment of the MeHg levels in fish
reflecting preparation. However, as discussed in section
1.4.4, close analysis of these studies resulted in a
conclusion that they in fact, did not support application
of alternative (lower) factors. Furthermore, given
available information, we do not believe that any
systematic bias is introduced into the analysis due to the
use of this parameter value.
Degree to which the potential impact of the source of
uncertainty is characterized as part of the analysis
Uncertainty in these factors was not quantitatively
assessed as part of the sensitivity analysis, however, as
noted in the cell to the left, the linear relationship between
both factors and exposure and risk allows us to readily
determine the magnitude of impact that uncertainty in
these factors could have on risk at the watershed-level.
Application of proportionality assumption in generating estimates of the U.S. EGU-attributable fraction of risk
(I) Fish tissue
measurement data
Concerns that the proportionality
assumption linking Hg deposition
If air deposition patterns from the 1 990s are reflected in
some of the Hg fish tissue measurements we are using
We did not quantitatively assess this potential source of
uncertainty and its impact on risk. However, our decision
100
-------
Source of
uncertainty
Description
Nature of potential impact on the exposure and risk
estimates
Degree to which the potential impact of the source of
uncertainty is characterized as part of the analysis
used in the analysis
may still reflect
earlier historical
patterns of Hg
deposition (from the
1990's) more so than
deposition from the
simulation period
(i.e., -2005)
and fish tissue Hg concentrations
may not hold if fish tissue Hg
concentrations actually reflect an
earlier (higher Hg deposition)
period. In this case, bias could be
introduced into estimates of
reduction in fish tissue Hg
concentrations (based on the
proportionality assumption) since
the effect of changes in current
Hg deposition on fish tissue Hg
concentrations could be
overstated (i.e., there could be an
effective buffering of the fish
tissue Hg concentrations by the
residual effects of earlier higher
Hg deposition).
the implications can vary depending on the nature of
that difference. If fish tissue levels for a watershed still
reflect higher 1990 deposition values, then there may be
an upward bias to the estimates of U.S. EGU-
attributable risk, since in reality we would expect the
underlying fish tissue levels to decrease as the impact of
those earlier higher deposition values dissipates.
Conversely, if total deposition remains the same, but
only the source distribution has changed since the
1990's, then the effect on our risk assessment may not
be that significant, since we are making projections
based on the current source-mix (or future source-mix)
assuming near steady-state assumptions are reached
given those specific source mixes. Given currently
available information, we cannot evaluate the interplay
of these two factors and the way in which they come
together to introduce potential bias into estimates of
U.S. EGU-attributable risk (we can only discuss each as
a source of uncertainty as has been done here).
to use only fish tissue samples from after 1999 is intended
to reduce this source of uncertainty.
(J) Watersheds can
display substantially
different methylation
potentials, resulting in
differences in the
impact of a unit
change in Hg
deposition on fish
tissue Hg
concentrations
Differences in methylation
potential cause variation in fish
tissue Hg concentrations across
waterbodies (even when those
waterbodies have similar rates of
Hg loading). In addition,
differences in methylation
potential can also translate into
different temporal profiles for the
response of fish tissue Hg
concentrations to changes in Hg
loading. Failure to reflect these
sources of variability in
methylation potential can
introduce uncertainty into the
analysis.
Differences in methylation potential across watersheds,
when present, are likely to be reflected in the underlying
fish tissue Hg concentrations themselves. Therefore,
these methylation differences will not to have a
substantial impact on our analysis and are not expected
to introduce systematic bias (the time sequence of
changes in fish tissue levels will depend on differences
in methylation potentials, but we are not attempting to
predict these temporal profiles).
We did not quantitatively assess uncertainty related to this
issue. Our determination is that this is not an important
source of uncertainty for our risk estimates.
(K) Factors related to
methylation in
watersheds (e.g.,
sulfate deposition,
pH) have not
remained constant
This source of uncertainty is
distinct from the issue of
methylation potential discussed
above (here we talk about factors
which can interact with the
underlying methylation potential
As with methylation potential discussed above, any
changes in factors affecting the underlying methylation
potential of a waterbody would eventually be reflected
in fish tissue Hg concentrations and therefore not
introduce systematic bias into the analysis. For this
reason, our concern over this issue is reduced. However,
We did not quantitatively assess uncertainty related to this
101
-------
Source of
uncertainty
Description
Nature of potential impact on the exposure and risk
estimates
Degree to which the potential impact of the source of
uncertainty is characterized as part of the analysis
over time, resulting in
variation in the
methylation potential
of watersheds over
time.
to increase or decrease it). If these
factors (pH and sulfate deposition
for example), have changed for a
given watershed, then the
methylation rate of that waterbody
may also change, thereby
impacting the degree to which
fish bioconcentrate and ultimately
(for higher trophic levels)
bioaccumulate Hg. Changes in
methylation factors can also
impact the temporal profile for
changes in fish tissue Hg
concentrations.
if changes in methylation factors are continuous and
extend into the current simulation period, then there is
more concern, since there could be more of a gradual
change in methylation across our simulation period (a
change that would not be fully reflected in the fish
tissue Hg values we have in the dataset). However, it is
difficult to characterize the magnitude of the potential
impact of this source of uncertainty on risk estimates,
including the direction and magnitude of any bias, given
the complex interplay of the factors involved (i.e.,
potential for ongoing changes in pH and sulfate to
produce methylation rates that vary over time).
(L) Effort to exclude
watersheds with
substantial non-air Hg
deposition58
Despite efforts to exclude
watersheds with substantial non-
air sources of Hg loading (see
section 1.4.2), some watersheds
with substantial non-air impacts
may have been retained in the
analysis.
The potential that we may have failed to exclude
watersheds with significant non-air Hg loadings could
introduce high-bias into our estimates of U.S. EGU-
attributable risk, since we would overstate the role of
U.S. EGUs in contributing to risk, by overlooking the
other non-air sources.
We completed two sensitivity analyses exploring this issue
and its potential impact on risk including: (a) an analysis
of risks when watersheds falling in LA, SC, MN, and ME,
are excluded (there is concern that these four states may
have significant non-air Hg releases and/or increased
methylation potential) and (b) an assessment of risk based
only on those watersheds falling in the upper 25th
percentile with regard to total Hg deposition (i.e.,
estimating risk when we focus on those watersheds where
we are more confident that aerial Hg deposition plays a
dominant role). The sensitivity analysis results suggest that
exclusion of the four states where we have concerns over
non-air Hg loading does not substantially impact risk.
However, when we estimated risk focusing on the
watersheds with the highest overall Hg loading (the upper
25th percentile), we did see a moderate increase in risk
estimates.
(M) Potential that the
modeling framework
is not sufficiently
refined from a spatial
standpoint to capture
elevated levels of Hg
If the overall approach for linking
Hg deposition (specifically the
fraction attributable to U.S.
EGUs) is not sufficiently spatially
refined, then areas of high U.S.
EGU Hg impact and resulting risk
We believe the level of uncertainty associated with
precision in modeling exposure and specifically the
ability to identify U.S. EGU-related risk "hot spots"
(again, using SAB terminology to refer to areas of
elevated Hg-related risk) is moderate. CMAQ modeling
is provided at the 12km grid resolution which matches
The issue of spatial scale in capturing high-end subsistence
fisher risk has not been quantitatively assessed as part of
the risk assessment.
58 For the March version of the Mercury Risk TSD, we also explored the issue of whether the proportionality assumption was more applicable to stationary
waterbodies (lakes/ponds) then flowing waterbodies (rivers/streams). However, for this version of the Mercury Risk TSD, reflecting recommendations made by
SAB that the proportionality assumption can be readily applied for either category of waterbody, we have not repeated that sensitivity analysis, or discussed this
source of uncertainty.
102
-------
Source of
uncertainty
Description
Nature of potential impact on the exposure and risk
estimates
Degree to which the potential impact of the source of
uncertainty is characterized as part of the analysis
deposition ("hot
spots" as termed in
the SAB comments)
(Note, this source of
uncertainty also
involves CMAQ
modeling)
may not be reflected in the risk
assessment. This issue speaks to
spatial precision in (a) the CMAQ
air quality modeling, (b) scale of
the watersheds (and degree to
which subsistence activity is
focused within specific
watersheds) and (c) applicability
of the proportionality assumption
in relating changes in Hg
deposition to changes in fish
tissue Hg concentrations (or
relating fraction of total Hg
deposition coming from U.S.
EGUs to the fraction of fish tissue
Hg attributable to U.S. EGUs).
well the watershed scale used in the analysis (i.e., the
HUC12). The HUC12 watersheds, in turn, represent a
small watershed which is appropriate if we assume that
the subsistence fisher being modeled focuses their
activity primarily at the same waterbody (i.e., within a
given HUC12 watershed). Therefore, given the focus of
this analysis on a reasonable highly-exposed
subsistence fisher (i.e., an individual who focuses their
activity at a given waterbody), then the modeling
framework, including the spatial scale of the various
modeling elements (CMAQ modeling, watershed, fish
tissue data), would seem appropriate. For this reason,
we believe there is little concern for the introduction of
bias into the risk assessment due to the precision of the
modeling and our ability to capture potential hot spots.
Factors relating to the estimation of mercury deposition over watersheds using the CMAQ model (e.g., estimating Hg emissions from U.S. EGUs and other sources, chemistry
associated with Hg fate and transport, prediction of wet and dry deposition, and global inflow of Hg into the U.S.)
(N) Hg 2005 U.S.
EGU emissions,
uncertainties and
sources of bias
2005 Hg emissions from EGUs
used for this analysis are from the
2005 inventory developed for the
National Air Toxics Assessment
(NATA). These data were
calculated by EPA by scaling Hg
data from the 1999 National
Emissions Inventory (NET) using
the ratio of 2005 heat input to
1999 heat input. Thus,
uncertainties and bias in the 2005
data fall into two classes - those
that are due to the 1999 estimates
and those resulting from the
scaling approach used to estimate
2005 emissions.
The 1999 NEI data uncertainties
include uncertainties traditionally
associated with emissions
estimation, such as uncertainties
in the accuracy of test data, the
applicability of the test data, the
representativeness of emission
The approach taken for the 1999 NEI used estimates
developed from the data collected for the ICR done for
the Clean Air Mercury Rule. These data were
preferentially used over other estimates available from
states, local agencies, or tribes. While these approaches
include some uncertainties, they used best available
practices and data available from the very rich data
source of the 1999 ICR. In general, emissions estimates
relying on test data will result in the best available
emission estimates for sources with such data, and the
uncertainties associated with measurement error are
considered negligible in comparison to the other sources
of uncertainty. Estimates based on emission factors will
have greater uncertainty but will not significantly
contribute to bias unless the emission factors are not
representative. Care was taken to apply the most
representative emission factors data to the sources in the
cases where this was needed. A great deal was known
about throughput at these sources from the ICR data,
and so that is not considered a large source of
uncertainty. These uncertainties can cause both
overestimates and underestimates of emissions, but
since there is no way to assess that, it is not possible to
quantify the potential impact (including the direction
The impact of this source of uncertainty was not
quantitatively evaluated.
103
-------
Source of
uncertainty
Description
Nature of potential impact on the exposure and risk
estimates
Degree to which the potential impact of the source of
uncertainty is characterized as part of the analysis
factors for specific units, the
oversimplification occurring when
emission factors are used rather
than continuous monitoring, and
the accuracy of throughput
estimates.
The 2005 approach adds
uncertainty as well as bias to
emissions estimates because it
does not include the impacts of
controls added to sources between
1999 and 2005 or changes in
fuels, including the Hg content of
the fuel. In addition, uncertainties
in the throughput information
used to scale the emissions can
affect the emissions estimates.
Lastly, the temporal and
speciation representation of
emissions in the 2005 model
inputs is another source of
uncertainty. However, as noted in
column to right, the net effect of
high-bias in 2005 emissions
would be to low-bias risk
estimates for the 2016 scenario.
and magnitude of bias) on risk estimates.
The 2005 scaling approach will tend to overestimate
emissions in 2005 because of additions of control
technologies and coal used between 1999 and 2005 that
would have reduced emissions. These reductions were
not included because only the changes to throughput
were considered in the approach. This most likely
causes uncertainties that result in a high bias
(overestimate) for the emissions in 2005. Since the
2005 deposition results are used in the denominator of
the risk scaling approach to predict risk in 2016,
overestimates of 2005 emissions are expected to result
in an underestimate of risk in 2016. The uncertainties
associated with the heat input data in 2005 are very low,
since most units included in the mercury estimates had
CEM data to record and report heat input for 2005.
The speciation approach used test data from the 1999
data collection, which assigned speciation factors based
on boiler type, fuel type, and control configuration.
While imperfect, these data will tend to minimize
inaccuracies in speciation assigned since the dataset
used to calculate the fractions was fairly large, and was
certainly the most comprehensive Hg speciation data
available. Finally, the temporal allocation approach was
kept the same between 2005 and 2016 so that any small
impacts of these assumptions on deposition results will
cancel out using the scaling approach for risk.
(0)Hg 2016 U.S.
EGU emissions,
uncertainties and
sources of bias
There are three key variables that
figure in the modeling of 2016 Hg
emissions from U.S. electric
generating units (U.S. EGUs): the
mercury content of the fuel
combusted (essentially various
ranks and grades of coal), the
extent of coal consumed, and type
and performance characteristics of
the controls to reduce mercury
emissions at a given U.S. EGU.
Uncertainty may figure in each of
these key variables.
As a result of uncertainty, the 2016 mercury emissions
from EGUs may be over or under-estimated. However,
the extent of the impact of uncertainty is considerably
reduced (as is concern over bias that may impact risk
estimates) due to the number of directly measured and
quality assured data points that are used to characterize
the variables described in the previous column, by the
mature nature of the technologies for mercury emission
control, and by the commensurate experience and
expertise in characterizing their performance as a result
of more than a decade of in-field operation of the
controls.
The impact of this source of uncertainty was not
quantitatively evaluated.
104
-------
Source of
uncertainty
Description
Nature of potential impact on the exposure and risk
estimates
Degree to which the potential impact of the source of
uncertainty is characterized as part of the analysis
(P) Hg Non-EGU
emissions, general
uncertainties and
consideration of
potential bias
The 2005 NATA inventory for
non-EGU sources is a compilation
of state data and data associated
with sector-specific rulemaking
activities at EPA. Uncertainties
include uncertainties traditionally
associated with emissions
estimation, such as uncertainties
in the accuracy of test data, the
applicability of the test data, the
representativeness of emission
factors for specific units, the
oversimplification occurring when
emission factors are used rather
than continuous monitoring, and
the accuracy of throughput
estimates. In general, the data
based on source tests and
individual source reporting are of
higher quality with less
uncertainty. In addition,
speciation and temporal allocation
considerations add sources of
uncertainty. These uncertainties
are also applicable to the 2016
emissions, since these emissions
are derived by scaling the 2005
emissions based on expected
reductions between 2005 and
2016.
Mercury emissions based on facility test data have been
used whenever possible for the 2005 NATA inventory,
focused on the categories with the highest emissions.
For example, emissions sources with 2 or more tons in
2005 include Portland cement, electric arc furnaces,
boilers, chemical manufacturing, hazardous waste
incineration, chlor-alkali plants, gold mining, and
municipal waste combustors. For these sources, which
represent 36 tons out of 47 tons non-EGU Hg
emissions, 65% of these sources used data developed
for rules associated with these sectors. The rule data
arein large part based on test data and have otherwise
had extensive review by industry experts and the
facilities themselves. In combination with the U.S.
EGU data, emissions from these higher quality data
sources comprises 76% of the total anthropogenic U.S.
Hg inventory. Thus, EPA's approach reflects an
inventory with much lower uncertainty than if it had
been based solely on more generic emission factors.
Speciation data used for this analysis are highly
uncertain, but reflect best available information. Since
the analysis was not quantifying the impact of non-EGU
sources, the speciation of non-EGU sources tends to
contribute less to the impact on the analysis.
Additionally, while temporal allocation uncertainties are
also high, the temporal allocation approaches for each
sector are used consistently in 2005 and 2016, so their
impacts are lessened through the use of the scaling
approach.
There are no known sources of bias related to
estimation of emissions from non-EGU sources in this
analysis. Every attempt was made to use the best
available emissions data in 2005 and future year
projections to 2016.
However, a possible (but not likely) source of non-EGU
emissions underestimates is from natural sources.
Public comments on EPA's approach suggest that some
natural sources were not considered. However, EPA
does include natural elemental Hg emissions from
volcanoes, oceans, and land. Further estimates of
recycled Hg emissions from oceans and land are
The impact of this source of uncertainty was not
quantitatively evaluated.
105
-------
Source of
uncertainty
Description
Nature of potential impact on the exposure and risk
estimates
Degree to which the potential impact of the source of
uncertainty is characterized as part of the analysis
included. One source of uncertainty is the treatment of
Hg from fires. While EPA did not explicitly include
these emissions in the analysis as fires, the model
calibration approach used to estimate the recycled land
emissions captures such emissions (Seigneur, et. al.
2004). To the extent that there is any underestimation
bias, it would reduce the fractional contribution of U.S.
EGU sources, however, given the low likelihood of
these sources of bias, the impacts on risk are expected
to be small.
(Q)Hg Emissions,
changes in
uncertainties and bias
between 2005 and
2016
The sources of change between
2005 and 2016 emissions are the
U.S. EGU emissions and the
stationary non-EGU emissions.
All other emissions sources
(mobile and non-anthropogenic)
stayed constant between the two
years. For U.S. EGUs, different
sources of uncertainties exist in
the two different approaches for
2005 and 2016 emissions, as
described above. The possible
changes in uncertainties from
2005 to 2016 are caused by the
changes in methods.
For non-EGU sources,
uncertainties in 2016 emissions
can result from assumptions about
emissions reductions associated
with other EPA regulations as
well as predicting future activity
levels associated with economic
or other changes.
For EGUs, some types of uncertainties could be greater
in 2016 than in 2005 since the emissions are forecast to
the future, for example, the future expected electricity
demand and coal usage. 2005 emissions estimates have
different uncertainties and it is not possible to determine
without extensive uncertainty analysis which sources
are greater than others. As described above, a bias most
likely exists in the 2005 data that does not exist in the
2016 data, resulting in a possible underestimate of risk
from the deposition-based scaling approach used for
2016 risk estimates.
For non-EGUs, several source categories reduced
emissions from 2005 to 2016 as a result of other EPA
regulations. While the reductions are estimated, the
uncertainties are relatively low given the extensive
analyses done to develop the regulations and since
compliance with EPA emissions reductions
requirements tends to be very high. Uncertainties are
introduced because EPA does not include co-benefits of
reductions from controls for other pollutants such as
PM2 5, which might lead to overestimation of the 2016
non-EGU emissions, lessening the impacts of U.S.
EGUs, and therefore lowering the risk estimates from
EGUs in our analysis. Uncertainty is also introduced in
this analysis for non-EGU sources because the possible
growth or reduction in emissions associated with
economic changes is not included in the 2016 estimates.
However, it is not possible to determine the impact of
excluding such considerations since non-EGU industry-
specific forecasts are highly uncertain themselves, there
is a complex relationship between economic changes,
regulations on emissions, and new source requirements
The impact of this source of uncertainty was not
quantitatively evaluated.
106
-------
Source of
uncertainty
Description
Nature of potential impact on the exposure and risk
estimates
Degree to which the potential impact of the source of
uncertainty is characterized as part of the analysis
that can reduce emissions.
(R) Uncertainty in
location of 2016
emissions reductions
Uncertainties in the 2016 scenario
regarding the specific geographic
locations of reductions in EGU-
derived mercury deposition as a
fraction of total mercury
deposition
The upstream U.S. EGU sources of Hg emissions are
provided by EPA's power sector model, the Integrated
Planning Model (IPM). The uncertainty in the extent of
projected Hg emissions from these sources is treated
above (see entry "O"). Here the focus is on the question
of uncertainty in the location of the 2016 emissions
reductions from the standpoint of which EGUs are
projected to operate and so serve as sources of Hg
emissions. IPM is an extremely detailed bottom-up
deterministic linear programming representation of the
U.S. power sector. There is no stochastic component in
the model that would allow one to characterize the
likelihood that the location of EGU sources and the
extent of emissions from these sources are shifting.
Therefore, we are not able at this time, to evaluate the
potential for systematic bias being introduced into risk
estimates through this step in our modeling. However,
the model's rigorous adherence to economic and
engineering fundamentals and detailed representation of
all factors affecting power system operation are
designed to limit uncertainty and ensure that the
model's locational projections are reasonable.
The impact of this source of uncertainty was not
quantitatively evaluated.
(S) Global inflow of
mercury into the
continental United
States
There is considerable uncertainty
in the global emissions inventory
for mercury and given the long
residence time of elemental
mercury it is possible that inflow
into the modeling domain may
reflect deficiencies in the global
emissions inventory. Some studies
suggest global mercury emissions
would increase without control
implementation (Pacyna et al.,
2009; Streets et al., 2009). Recent
measurements of ambient
mercury at remote locations from
1975 to 2008 suggest global
mercury burden is steadily
decreasing (Slemr et al., 2011).
Given this conflicting information
and the uncertainty in projecting
Mercury deposition may be over or under-estimated
(i.e., potential bias introduced is not clear in terms of
direction or magnitude).
This analysis did not address this source of uncertainty.
Mercury inflow was adjusted in another study and was
shown to have a larger impact on modeled dry deposition
compared to wet deposition. Boundary condition
perturbations within realistic bounds resulted in minimally
changed distributions of total modeled mercury deposition
(Pongprueksa et al., 2008). This study also determined that
model spin-up of one week is necessary to remove
influence of initial conditions (Pongprueksa et al., 2008).
The CMAQ modeling in this analysis had spin-up of more
than one week making initial condition influence
negligible.
107
-------
Source of
uncertainty
Description
Nature of potential impact on the exposure and risk
estimates
Degree to which the potential impact of the source of
uncertainty is characterized as part of the analysis
the current global emission
inventory, boundary inflow is
kept constant between 2005 and
2016.
(T) Photochemical
model prediction of
mercury wet
deposition estimates
The uncertainty in the wide
variety of model inputs such as
emissions, meteorology, global
inflow to the modeling domain,
and chemistry manifest in model
estimates of mercury wet
deposition.
Mercury wet deposition may be under-estimated during
the summer season and over-estimated during the
winter (i.e., potential bias introduced is not clear in
terms of direction or magnitude). Model performance is
described in "Air Quality Modeling Technical Support
Document: Point Source Sector Rules" (USEPA,
201 Ib).
Model estimated weekly mercury wet deposition is
compared to observation data to assess model skill
simulating this component of mercury deposition. Mercury
wet deposition measurements are weekly totals taken at
sites that are part of the Mercury Deposition Network
(http://nadp.sws.uiuc.edu/MDN/) which operates under the
National Atmospheric Deposition Program. This is
generally consistent with other published studies that use
coarser grid resolution, older versions of CMAQ and older
emission inventories (Bullock et al., 2008; Lin et al.,
2007).
(U) Photochemical
model prediction of
mercury dry
deposition estimates
The uncertainty in the wide
variety of model inputs such as
emissions, meteorology, global
inflow to the modeling domain,
and chemistry manifest in model
estimates of mercury dry
deposition.
Mercury dry deposition may be over or under-estimated
(i.e., potential bias introduced is not clear in terms of
direction or magnitude).
This analysis did not address this source of uncertainty.
There is a lack of dry deposition observation data that
makes a direct quantitative or even qualitative comparison
to modeled estimates impossible. Other studies have
shown differences in estimated dry deposition based on
changes in ambient mercury, reaction rate changes, and
changes to the dry deposition scheme (Bullock et al., 2009;
Lin et al., 2007; Pongprueksa et al., 2008; Ryaboshapko et
al., 2007a; Ryaboshapko et al., 2007b). However, given
that these studies are based on earlier versions of CMAQ
or other models that are not routinely used for regulatory
purposes and that mercury chemistry and dry deposition
has changed through model versions, it is not clear that dry
deposition estimates in the version of CMAQ used for this
analysis would be comparable to earlier studies.
(V) Mercury
Chemistry
The complete set of mercury
oxidation and reduction reactions
has not been identified by the
scientific community.
Mercury deposition may be over or under-estimated
(i.e., potential bias introduced is not clear in terms of
direction or magnitude).
This analysis did not address this source of uncertainty.
Other studies have shown differences in estimated total
mercury deposition based on changes in reaction rates
(Bullock et al., 2009; Lin et al., 2007; Pongprueksa et al.,
2008; Ryaboshapko et al., 2007a; Ryaboshapko et al.,
2007b). However, given that these studies are based on
earlier versions of CMAQ or other models that are not
routinely used for regulatory purposes and that mercury
chemistry and deposition has changed through model
versions it is not clear that total mercury deposition
estimates in the version of CMAQ used for this analysis
would be comparable to earlier studies.
108
-------
Source of
uncertainty
Description
Nature of potential impact on the exposure and risk
estimates
Degree to which the potential impact of the source of
uncertainty is characterized as part of the analysis
Application of the MeHg RfD in generating hazard quotient (HQ) risk estimates
(W) Degree to which
the RfD as calculated
provides coverage for
low SES status
groups which may be
at greater risk for
adverse health effects
following MeHg
exposure
Given a number of factors (e.g.,
nutritional deficiencies, reduced
access to health care and health-
related information), there is
concern that the MeHg RfD may
not provide sufficient coverage
for these low SES populations.
The RfD is defined as the amount of the substance of
concern that can be consumed without expectation of
harm for a lifetime by populations including sensitive
subpopulations. The calculated RfD for MeHg includes
an uncertainty factor to account for human
pharmacokinetic variability (3 fold) and uncertainty
and variability in pharmacodynamics. This may be
sufficient to account for increased sensitivity to IQ
decrements or adverse effects on neurobehavioral
functions in low socio-economic status populations.
There are no published analyses by EPA or other parties
that would permit estimation of uncertainty for this
factor and the potential for resulting bias being
introduced into the analysis. Two of the human
populations on which the RfD was based (the Faroese
and the Seychellois) are relatively homogenous for
some aspects of SES.
We did not explore this potential source of uncertainty
quantitatively. However, as noted in the cell to the left due
to the method used in calculating the RfD, concerns that
the RfD may not provide coverage for low SES
populations is reduced.
(X) The MeHg RfD
was derived based on
saltwater fish
consumption (which
can involve higher
levels of nutrients that
ameliorate the
adverse effects of
MeHg). However, the
RfD is now being
applied in the context
of freshwater fish
consumption (which
can involve lower
levels of these
nutrients).
Recent studies (e.g. Oken et al
2008; Choi et al 2008;) point to
the potential for nutrients in fish
(particularly marine fish) to
ameliorate some of the observed
adverse effects of MeHg when co-
exposure occurs. However, there
was no correction for potential
confounding by nutrients in
marine fish and mammals in
calculation of the benchmark
doses used in the RfD derivation.
Therefore, there is the potential
that these benchmark doses may
be underestimates, in which case
the HQ estimates based on the
RfD could be biased low,
particularly in the case of
freshwater fish which may have
lower levels of these nutrients.
Failure to consider the effects offish nutrients as a
covariate or confounder for nerodevelopmental effects
associated with MeHg exposure could result in HQ
estimates being biased low. However, currently,
available information does not support a rigorous
adjustment of the RfD to address potential confounding
by fish nutrients.
We did not explore this potential source of uncertainty
quantitatively.
109
-------
3 Summary of Key Observations
This section provides key policy-relevant observations drawn from discussions presented
in Sections 2.1 through 2.7. It is important to emphasize, that the risk estimates and additional
supporting information summarized here are intended to inform a determination by the
Administrator as to whether Hg emitted from U.S. EGUs represents a public health hazard.
These observations are not intended to be conclusionary in nature and instead, focus on
characterizing the nature and magnitude of risk associated with U.S. EGU Hg emissions.
• Estimates of U.S. EGU Hg emissions suggest that the 2016 Scenario is likely closer to
recent (2010) emissions compared with the 2005 scenario (which has substantially higher
total Hg emissions for this sector). Therefore, risk estimates have been generated only for
the 2016 scenario for the revised risk assessment (CMAQ-based Hg deposition estimates
for the 2005 scenario are used in scaling fish tissue concentrations for use in modeling
2016 scenario risk).
• Risk characterization is based on estimates of RfD-based HQ. Due to concerns raised by
the SAB peer-review panel that the IQ loss endpoint may not fully capture the range of
neurodevelopmental effects associated with MeHg exposure, IQ loss estimates are
presented in brief summary form in Appendix B and not used in the risk characterization.
• Based on the 2016 scenario, U.S. EGUs can contribute up to 11% of total Hg emissions
for a subset of watersheds. However, in general, other sources besides U.S. EGUs
dominate Hg deposition. In 2016 U.S. EGUs contributed on average about 2% of total
Hg deposition across the country. U.S. EGU-related Hg deposition is higher in the eastern
part of the country with elevated contributions in a number of specific areas, including
most notably, the Ohio River valley. U.S. EGU-related Hg deposition estimates show a
significant reduction between 2005 and 2016 scenarios, reflecting mainly co-benefits
from implementation of criteria pollutant controls with the average U.S. EGU-attributable
deposition decreasing from -5% of total to -2% for the 2005 and 2016 scenarios,
respectively.
• Based on the 2016 scenario, U.S. EGUs can contribute up to 16% of MeHg in fish tissue.
However, generally, U.S. EGUs contribute a much smaller fraction averaging 3% for the
2016 Scenario.
• Comparing the magnitude of Hg fish tissue levels with total Hg deposition (as
characterized at the watershed-level) suggests that there is not a strong correlation. This is
not surprising given the variety of factors which effect methylation potential; factors
which can demonstrate substantial spatial variation. However, available evidence
supports a steady state linear relationship between changes in Hg deposition and changes
in fish tissue Hg concentrations.
• The additional fish tissue Hg data added for the revised risk assessment have significantly
improved coverage for watersheds located in areas with elevated U.S. EGU-related Hg
deposition such as the Ohio River Valley. However, we still conclude that, generally, our
110
-------
coverage for high U.S. EGU impact areas remains limited. For this reason, we continue to
believe that the actual number of "at-risk" watersheds (i.e., watersheds where U.S. EGUs
could contribute to a public health concern) could be substantially larger than estimated.
• Based on application of the 2-stage risk characterization framework described in Section
1.3, we estimate that from 22 to 29% of the watersheds included in this risk assessment
could be classified as potentially having at-risk populations under the 2016 Scenario.
This estimate is based on risks modeled for the typical female subsistence fish consumer
and reflects aggregation of results from Stages la and Ib of the 2-Stage Risk
Characterization Framework (i.e., watersheds where mercury released from U.S. EGUs
when considered alone, without taking into account mercury deposition from other
sources, would produce an HQ >1 or watersheds where total HQ (reflecting Hg from all
sources) is > 1 and U.S. EGUs make at least at 5% contribution to that-risk).
• Comparison of risk estimates generated for the typical female subsistence fish consumer
scenario with estimates generated for the other six SES-differentiated female subsistence
fish consumer scenarios included in this risk assessment results in the following
observations: (a) total and U.S. EGU-attributable risks for the Hispanic and Vietnamese
scenarios are generally lower than for the typical female subsistence fish consumer
scenario, (b) U.S. EGU-attributable risks for the Tribal scenario are similar to those for
the typical female subsistence fish consumer scenario, although total risks are generally
higher for the Tribal scenario and (c) U.S. EGU-attributable risks for the Laotian and low
income southeastern White and Black scenarios are generally higher than for the typical
female subsistence fish consumer scenario, although total risks can be higher or lower
depending on the scenario. These trends suggest that generally, the typical female
subsistence fish consumer scenario will provide coverage (in terms of representing risk)
for Hispanic, Vietnamese and Tribal scenarios. However, the typical female subsistence
fish consumer scenario may not provide full coverage for the Laotian and low income
southeastern White and Black scenarios, particularly in terms of U.S. EGU-attributable
risk.
• If U.S. EGU impacts to watersheds included in the risk assessment were zeroed-out, for a
significant majority of those watersheds, total exposure would still exceed (and in most
cases, significantly exceed) the RfD. Reductions in U.S. EGU attributable Hg will
reduce the magnitude of the risk, although substantial total exposure and risk from Hg
deposition will remain.
• Sensitivity analyses conducted primarily to examine uncertainty in applying the
proportionality assumption linking Hg deposition to Hg fish tissue levels, suggest that
uncertainty related to the proportionality assumption is unlikely to substantially effect an
assessment of whether Hg emissions from U.S. EGUs constitute a public health concern.
Use of watershed-level 50th percentile fish tissue Hg concentrations (instead of the 75th
percentile values used in the core analysis) can result in notable reductions in risk
estimates in some instances, but the SAB peer review panel supports use of the 75th
percentile estimates.
Ill
-------
Citations:
Appel, K. W., A. B. Gilliland, et al. (2007). "Evaluation of the Community Multiscale Air
Quality (CMAQ) model version 4.5: Sensitivities impacting model performance Part I -
Ozone." Atmospheric Environment 41(40): 9603-9615.
Appel, K. W., P. V. Bhave, et al. (2008). "Evaluation of the community multiscale air quality
(CMAQ) model version 4.5: Sensitivities impacting model performance; Part II -
particulate matter." Atmospheric Environment 42(24): 6057-6066.
Axelrad, D. A.; Bellinger, D. C.; Ryan, L. M.; Woodruff, T. J. (2007). Dose-response
relationship of prenatal mercury exposure and IQ: an integrative analysis of
epidemiologic data. Environmental Health Perspectives. 2007, 115, 609-615.
Bloom, N.S. (1992). On the chemical form of mercury in edible fish and marine invertebrate
tissue. Can. J. Fish. Aquat. Sci. 49:1010-1017.
Bullock, O. R. and K. A. Brehme (2002). "Atmospheric mercury simulation using the CMAQ
model: formulation description and analysis of wet deposition results." Atmospheric
Environment 36(13): 2135-2146.
Bullock, O.R., Atkinson, D., Braverman, T., Civerolo, K., Dastoor, A., Davignon, D., Ku, J.Y.,
Lohman, K., Myers, T.C., Park, R.J., Seigneur, C., Selin, N.E., Sistla, G.,
Vijayaraghavan, K., (2008). The North American Mercury Model Intercomparison Study
(NAMMIS): Study description and model-to-model comparisons. Journal of Geophysical
Research-Atmospheres 113.
Bullock, O.R., Atkinson, D., Braverman, T., Civerolo, K., Dastoor, A., Davignon, D., Ku, J.Y.,
Lohman, K., Myers, T.C., Park, R.J., Seigneur, C., Selin, N.E., Sistla, G.,
Vijayaraghavan, K., (2009). An analysis of simulated wet deposition of mercury from the
North American Mercury Model Intercomparison Study. Journal of Geophysical
Research-Atmospheres 114.
Burger, J. (2002). Daily consumption of wild fish and game: Exposures of high end
recreationalists, International Journal of Environmental Health Research, 12:4, p. 343-
354, July, 2010.
Burger, J., Stephens, W. L., Boring, C. S., Kuklinski, M., Gibbons, J. W., Gochfeld M. (1999)
Factors in Exposure Assessment: Ethnic and Socioeconomic Differences in Fishing and
Soncumption of Fish Caught along the Savannah River. Risk Analysis, Vol. 19, No. 3, p.
427.
Byun, D. and K. L. Schere (2005). "Review of the governing equations, computational
algorithms, and other components of the models-3 Community Multiscale Air Quality
(CMAQ) modeling system." Applied Mechanics Reviews 59(1-6): 51-77'.
Byun, D., Schere, K.L., 2006. Review of the governing equations, computational algorithms, and
other components of the models-3 Community Multiscale Air Quality (CMAQ) modeling
system. Applied Mechanics Reviews 59, 51-77'.
Dellinger, J. A., (2004) Exposure assessment and initial intervention regarding fish consumption
of tribal members in the Upper Great Lakes Region in the United States. Dellinger,
Environmental Research 95, p. 325-340
112
-------
Dispasquale, M.M., J. Agee, C. McGowan, R.S. Oremland, M. Thomas, D. Krabbenhoft, and
C.C. Gilmour. (2000) Methyl-Mercury Degradation Pathways: A Comparison Among
Three Mercury-Impacted Ecosystems. Environ. Sci. Technol., 34, 4908-4916.
Farias L.A., Favaro, D.I., Santos J.O., Vasconcellos M.B., et al., (2010). Cooking Process
Evaluation on Mercury Content in Fish,|| Acta Amazonia, 40 (4), 741-748.
Grandjean, P., P. Weihe, R. White, F. Debes, S. Arak, K. Yokoyama, K. Murata, N. Sorensen, R.
Dahl, and P. Jorgensen. (1997). Cognitive deficit in 7-year-old children with prenatal
exposure to methylmercury. Neurotoxicol. Teratol. 20:1-12.
Harris., R. C., John W. M. Rudd, Marc Amyot, Christopher L. Babiarz, Ken G. Beaty, Paul J.
Blanchfield, R. A. Bodaly, Brian A. Branfireun, Cynthia C. Gilmour, Jennifer A.
Graydon, Andrew Heyes, Holger Hintelmann, James P. Hurley, Carol A. Kelly, David P.
Krabbenhoft, Steve E. Lindberg, Robert P. Mason, Michael J. Paterson, Cheryl L.
Podemski, Art Robinson, Ken A. Sandilands, George R. Southworth, Vincent L. St.
Louis, and Michael T. TateRudd, J. W. M., Amyot M., et al., (2007). Whole-Ecosystem
study Shows Rapid Fish-Mercury Response to Changes in Mercury Deposition.
Proceedings of the National Academy of Sciences Early Edition, PNAS 2007 104 (42)
pp. 16586-16591; (published ahead of print September 27, 2007).
Knights, D. C, Sunderland, E. M., Barber, M. C., Johnston J. M., and Ambrose, R. B., (2009).
Application of Ecosystem-Scale Fate and Bioaccumulation Models to Predict Fish
Mercury Response Times to Changes in Atmospheric Deposition. Environmental
Toxicology and Chemistry, Vol 28, No. 4., pp. 881-893.
Lin, C.J., Pongprueks, P., Rusell Bulock, O., Lindberg, S.E., Pehkonen, S.O., Jang, C.,
Braverman, T., Ho, T.C., (2007). Scientific uncertainties in atmospheric mercury models
II: Sensitivity analysis in the CONUS domain. Atmospheric Environment 41, 6544-6560.
Morgan, J.N., M.R. Berry, and R.L. Graves. (1997). "Effects of Commonly Used Cooking
Practices on Total Mercury Concentration in Fish and Their Impact on Exposure
Assessments." Journal of Exposure Analysis and Environmental Epidemiology 7(1): 119-
133.
Morgan, J.N., M.R. Berry, Jr., and R.L. Graves. (1994). Effects of Native American cooking
practices on total mercury concentrations in walleye. Presented at ISEE/ISEA Joint
Conference, September 18-21, 1994.
Moya. J., Itkiin, C., Selevan, S.G., Rogers, J.W., Clickner, R. P., (2008). Estimates offish
consumption rates for consumers of bought and self-caught fish in Connecticut, Florida,
Minnesota, and North Dacota. Science of the Total Environment (2008),
doi:10.1016/j.scitotenv.2008.05.023.
Musaiger, A. O., R. D'Souza, (2008). The effects of different methods of cooking on proximate,
mineral and heavy metal composition offish and shrimps consumed in the Arabian Gulf.
Archives LatinoAmericanos De Nutricion, Organo Oficial de la Sociedad
Latinoamericana de Nutricion., Vol 58, No 1, 2008.
Oken e., Radesky, J.S., Wright, R. O., Bellinger D. C., Amarasiriwardena C. UJ., Kleinman, K.
P., Hu, H., Gillman, M.W., (2008). Maternal Fish Intake during Pregnancy, Blood
113
-------
Mercury Levels, and Child Cognition at Age 3 Years in a US Cohort. American Journal
of Epmidemiology, Vol. 167, No. 10, p. 1171.
Pacyna, E.G., Pacyna, J.M., Sundseth, K., Munthe, J., Kindbom, K., Wilson, S., Steenhuisen, F.,
Maxson, P., 2009. Global emission of mercury to the atmosphere from anthropogenic
sources in 2005 and projections to 2020. Atmospheric Environment 44, 2487-2499.
Pirrone N. and Keating T, editors. "Hemispheric Transport of Air Pollution (2010), Part B:
Mercury. Prepared by the Task Force on Hemispheric Transport of Air Pollution acting
within the framework of the Convention on Long -range Transboundary Air Pollution.
United Nations New York and Geneva, 2010.
Pirrone N, Cinnirella S, Feng X, Finkelman RB, Friedli HR, Leaner J, et al., (2010). Global
mercury emissions to the atmosphere from anthropogenic and natural sources.
Atmospheric Chemistry and Physics Discussions, 10(2), 4719-4752.
Pongprueksa, P., Lin, C.J., Lindberg, S.E., Jang, C., Braverman, T., Bullock, O.R., Ho, T.C.,
Chu, H.W., (2008). Scientific uncertainties in atmospheric mercury models III: Boundary
and initial conditions, model grid resolution, and Hg(II) reduction mechanism.
Atmospheric Environment 42, 1828-1845.
Rice GE, Hammitt JK, Evans JS. (2010). A probabilistic characterization of the health benefits of
reducing methyl mercury intake in the United States. Environmental Science Technology.
2010 Jul l;44(13):5216-24
Ryaboshapko, A., Bullock, O.R., Christensen, J., Cohen, M., Dastoor, A., Ilyin, I, Petersen, G.,
Syrakov, D., Artz, R.S., Davignon, D., Draxler, R.R., Munthe, J., (2007a).
Intercomparison study of atmospheric mercury models: 1. Comparison of models with
short-term measurements. Science of the Total Environment 376, 228-240.
Ryaboshapko, A., Bullock, O.R., Christensen, J., Cohen, M., Dastoor, A., Ilyin, I, Petersen, G.,
Syrakov, D., Travnikov, O., Artz, R.S., Davignon, D., Draxler, R.R., Munthe, J., Pacyna,
J., (2007b). Intercomparison study of atmospheric mercury models: 2. Modelling results
vs. long-term observations and comparison of country deposition budgets. Science of the
Total Environment 377, 319-333.
Seigneur, C., K . Vigjayaraghavan, K. Lohman, P. Karamchandani, C. Scott. (2004)."Global
Source Attribution for Mercury Deposition in the United States." Environ. Sci. Technol.
38, 555-569.
Selin, N.E., Jacob, D.J., Park, R.J., Yantosca, R.M., Strode, S., Jaegle, L., Jaffe, D., (2007).
Chemical cycling and deposition of atmospheric mercury: Global constraints from
observations. Journal of Geophysical Research-Atmospheres 112.
Shilling, Fraser, Aubrey White, Lucas Lippert, Mark Lubell (2010). Contaminated fish
consumption in California's Central Valley Delta. Environmental Research 110, p. 334-
344.
Slemr, F., Brunke, E.G., Ebinghaus, R., Kuss, J., (2011). Worldwide trend of atmospheric
mercury since 1995. Atmospheric Chemistry and Physics 11, 4779-4787.
Streets, D.G., Zhang, Q., Wu, Y., (2009). Projections of Global Mercury Emissions in 2050.
Environmental Science & Technology 43, 2983-2988.
114
-------
Swartout, J., and G. Rice. (2000). "Uncertainty Analysis of the Estimated Ingestion Rates Used
to Derive the Methylmercury Reference Dose." Drug and Chemical Toxicology
23(1):293-306. 11-41
U.S. EPA, U.S. Environmental Protection Agency, (1997). Mercury Study Report to Congress.
Office of Air Quality Planning and Standards and Office of Research and Development.
EPA-452/R-97-003, December 1997.
U.S. EPA, U.S. Environmental Protection Agency, (2000). Guidance for Assessing Chemical
Contaminant Data for Use in Fish Advisories, Volume 1: Fish Sampling and Analysis
and Volume 3: Overview of Risk Management. Office of Science and Technology,
Office of Water, U.S. Environmental Protection Agency, Washington, DC. EPA 823-B-
00-007.
U.S. EPA, U.S. Environmental Protection Agency, (200la). Water Quality Criterion for the
Protection of Human Health: Methylmercury - Final. U.S. EPA, Office of Science and
Technology, Office of Water, EPA-823-R-01-001 January 2001.
U.S. EPA, U.S. Environmental Protection Agency, (2001b). Mercury Maps - A Quantitative
Spatial Link Between Air Deposition and Fish Tissue Peer Reviewed Final Report. U.S.
EPA, Office of Water, EPA-823-R-01-009, September, 2001.
U.S. EPA, U.S. Environmental Protection Agency, (2005), Neurobehavioral Assessments
Conducted in the New Zealand, Faroe Islands, and Seychelles Islands Studies of
Methylmercury Neurotoxicity in Children. David C. Bellinger, Professor of Neurology,
Harvard Medical School, March 2005. Report to the U.S. Environmental Protection
Agency EPA-HQ-OAR-2002-0056-6045[ 1 ]
U.S. EPA, U.S. Environmental Protection Agency, (2007a) Review of the National Ambient Air
Quality Standards for Lead: Policy Assessment of Scientific and Technical Information
OAQPS Staff Paper. Office of Air Quality Planning and Standards, Research Triangle
Park, NC. EPA-452/R-07-013 November 2007.
U.S. EPA, U.S. Environmental Protection Agency, (2007b) Lead: Human Exposure and Health
Risk Assessments for Selected Case Studies, Volume I. Human Exposure and Health
Risk Assessments—Full-Scale and Volume II. Appendices. Office of Air Quality
Planning and Standards, Research Triangle Park, NC. EPA-452/R-07-014a and EPA-
452/R-07-014b.
U.S. EPA, U.S. Environmental Protection Agency, (201 la). Technical Support Document:
National-Scale Mercury Risk Assessment Supporting the Appropriate and Necessary
Finding for Coal- and Oil-Fired Electric Generating Units. March, 2011. EPA-452/D-11-
002.
U.S. EPA, U.S. Environmental Protection Agency, (201 Ib). Air Quality Modeling Technical
Support Document: EGU Mercury Analysis. EPA-454/R-11-008.
UNEP's Division of Technology, Industry and Economics. (2010). Study on mercury sources
and emissions and analysis of cost and effectiveness of control measures "UNEP
Paragraph 29 study".
115
-------
USGS, U.S. Geological Survey (USGS). 2005. Active mines and mineral plants in the US.
Reston, VA. Available on the Internet at http://tin.er.usgs.gov/mineplant
USGS, U.S. Geological Survey and U.S. Department of Agriculture, Natural Resources
Conservation Service, (2009), Federal guidelines, requirements, and procedures for the
national Watershed Boundary Dataset: U.S. Geological Survey Techniques and Methods
11-A3, 55 p.
Wijngaarden, Edwin van, Christopher Beck, Conrad F. Shamlaye, Elsa Cernichiari, Philip W.
Davidson, Gary J. Myers, Thomas W. Clarkson. (2006). Benchmark concentrations for
methyl mercury obtained from the 9-year follow-up of the Seychelles Child Development
Study. NeuroToxicology 27 :702-709.
116
-------
Appendices: Additional Technical Detail on Modeling Elements and Presentation of
Supplemental Risk Estimates
(Citations for the appendices are provided in the citation list for the main document -
see above)
As noted in section 1.4.5, the IQ loss risk metric has been de-emphasized in presenting
risk estimates due to concerns that it may not fully capture the range of adverse
neurodevelopmental effects associated with MeHg exposure. For this reason, both the
description of the approach used in modeling IQ loss as well as a summary of IQ loss estimates
and a discussion of uncertainty related to those risk estimates is presented in appendices, with the
technical approach being described in Appendix A and the summary of IQ loss risk estimates
and discussion of uncertainty being presented in Appendix B. Appendix C presents the SAB
Mercury Panel peer review letter (the attachment to that letter provides the original EPA charge
questions).
Appendix A. Technical Approach Used in Modeling IQ Loss
Estimation of IQ loss in children begin with the same exposure estimates used in
generating HQ estimates (i.e., estimates of MeHg exposure generated for a given subsitence
fisher scenario - see section 1.4.4 for derivation of these exposure estimates). However, these
estimates of body weighted-adjusted MeHg exposure need to be converted into an equivalent
maternal hair concentration since the IQ loss function uses hair Hg as the dose measure. To do
that, we use a dose-to-hair conversion factor (DHCV) of 12.5 (units ppm per unit |ig/kg-day) that
converts ingested dose (IR) to hair Hg concentration in ppm. The DHCV factor is based on a one
compartment toxicokinetic model used for deriving the MeHg RfD by Swartout and Rice (2000).
After generating an estimate of maternal hair Hg level for the subsistence fisher (at the
particular watershed being modeled), we then apply a concentration-response (CR) function
relating material hair Hg levels to IQ points lost in the child born to that mother. This CR
function was published in Axelrad et al., 2007and is based on application of a Bayesian
hierarchical model which integrates data from the three key epidemiological studies (Seychelles,
New Zealand and Faroe Islands).59
Since the CR function was published in the Axelrad et al., 2007 study, a number of
authors have raised the possibility that neurological deficits related to Hg exposure through fish
consumption could be masked to some degree by the neurologically-beneficial effects offish oil
consumption. Some authors have suggested that the IQ loss factor should be adjusted upward to
compensate for this masking effect (see Rice et al., 2010 and Oken, 2008). However, no rigorous
basis for a specific adjusted estimate has been provided to-date and therefore, we address this
potential for low-bias as part of our qualitative uncertainty discussion (see Table B-2, Appendix
B).
59 The IQ loss model uses a linear slope of 0.18 IQ points per ppm hair Hg concentration (Axelrad et al., 2007).
117
-------
Appendix B. Supplemental Risk Estimates (IQ loss estimates)
This section provides summaries of IQ loss risk estimates generated for the 2016 scenario
and discusses key sources of uncertainty associated with the IQ loss estimates. In assessing the
potential public health significance of the IQ loss risk estimates, based on recommendations
provided by the Clean Air Science Advisory Committee (CAS AC) in the context of the last
National Ambient Air Quality Standard (NAAQS) review for lead completed in 2008 (US EPA,
2007a), we interpreted IQ loss estimates of 1-2 points as being clearly of public health
significance. All of the risk estimates summarized here are based on application of the revised
version of the risk assessment model. Specific tables include:
• Table B-l: provides risk percentiles for the IQ loss risk metric for children born to
members of the typical female subsistence fish consumer scenario for the 2016 scenario.
• Table B-2: identifies and discusses key sources of uncertainty associated with modeling
IQ loss (uncertainty related to modeling exposure, which also impacts HQ estimates, is
covered in Table 2-15).
Table B-l. Percentile IQ loss risk estimates for children born to members of the typical
female subsistence fish consumer scenario assessed nationally (2016 scenario) (for
both total and U.S. EGU incremental risk, including IQ loss and MeHg RfD-based HQ
estimates)
Typical female
subsistence fish
consumer rate
(g/day) and
percentile/mean
39 (mean)
123 (90th)
173 (95th)
373 (99th)
Watershed percentile
Total IQ points lost
50th
0.4
1.4
1.9
4.1
75th
0.8
2.4
3.4
7.4*
90th
1.3
4
5.6
12.1*
95th
1.7
5.3
7.4*
16*
99th
2.5
8*
11.2*
24.1*
U.S. EGU-attributable IQ points lost
50th
-
-
-
0.1
75th
-
0.1
0.1
0.2
90th
-
0.1
0.2
0.3
95th
0.1
0.2
0.2
0.5
99th
0.1
0.3
0.4
0.8
-IQ loss is <0.1 point
* IQ loss estimate subject to greater uncertainty due to application of the underlying concentration-response function
for IQ loss at levels of exposure above those in the underlying epidemiological studies (see Appendix A)
Table B-2. Key sources of uncertainty related to modeling IQ loss, the nature of their
potential impact on risk estimates, and the degree to which they are characterized as
part of the analysis
Source of
uncertainty
(A) IQ may not
fully capture
the most
Description
IQ may not represent the most
sensitive cognitive endpoint for
Hg exposure (Axelrad et al.,
Nature of potential impact on
the exposure and risk estimates
Given this concern, we have
focused the risk assessment on the
RfD-based HQ risk estimates and
Degree to which the potential
impact of the source of
uncertainty is characterized as
part of the analysis
Because we do not have readily
available data to support
quantitative analyses of the first
118
-------
Source of
uncertainty
Description
Nature of potential impact on
the exposure and risk estimates
Degree to which the potential
impact of the source of
uncertainty is characterized as
part of the analysis
sensitive
cognitive
endpoints
associated with
Hg exposure
2007-see Section 1.2). In
addition deficits in some
categories of cognitive
functioning are not captured by
IQ. Together, these sources of
uncertainty suggest that we
could be under-estimating the
extent of cognitive impacts
associated with Hg exposure
(when we focus on modeling the
IQ loss endpoint alone).
have deemphasized IQ loss
estimates.
two sources of uncertainty (IQ
loss not capturing all of the
cognitive effects and potential
confounding by LCPFAs), we
could only address these factors
qualitatively. In both cases, the
potential effect on the risk
assessment would be to
potentially down-bias our
estimate of cognitive endpoint-
related risk for children.
(B) Potential
confounding
from nutrients,
including long-
chained
polyunsaturated
fatty acids
(LCPFAs),
found in fish
This issue is similar to the
nutrient confounding issue
discussed above under the RfD-
based HQ entry. Specifically,
there are concerns that fish
nutrients may mask to some
extent, the adverse
neruodevelopmental effects of
MeHg, resulting in a
concentration-response function
for IQ loss which could be
biased low (particularly in the
case of freshwater fish
consumption which may have
lower levels of relevant fish
nutrients)
Potential confounding (masking)
of the neurodevelopmental effects
of MeHg could result in
concentration-response functions
for IQ loss that are low-biased,
which in turn would mean that
estimates of IQ loss could be low
biased. However, at this point we
do not have the information
necessary to derive a
concentration-response function
for IQ loss that would account for
this masking effect and therefore,
cannot quantitatively evaluate this
source of uncertainty.
(C) How to
treat potential
outliers from
the
epidemiological
datasets used in
deriving the IQ
loss functions.
Regarding outliers, when an
outlier datapoint from the
Seychelles study was included in
the integrated derivation of the
IQ loss slope factor, the factor
was reduced by 25 percent (from
-0.18 IQ points per unit ppm hair
Hg, to -0.125). If in reality, this
outlier actually reflects the true
response for a subset of the
populations, then risks (as
modeled) could be biased high
specifically for this
subpopulation.
In the case of excluding the
outlier from the Seychelles study,
we note that the effect (given the
linear nature of the IQ loss slope)
would be to simply result in a
25% reduction in risk, if we were
to include the outlier in derivation
of the slope function (i.e., a
formal rerun of the model with
this alternative slope is not
required - we can just consider
this magnitude of impact on the
primary risk estimates we
generate for the analysis).
Although it is possible given the
linear relationship of exposure
and IQ loss to readily determine
the impact of a sensitivity
analysis involving application of
an IQ loss slope factor that
reflects inclusion of the outliers,
we did not explicitly complete
such an analysis (primarily
because the IQ loss risk metric
has been de-emphasized with
primary focus being placed on
the RfD-based HQ metric).
119
-------
Appendix C. SAB Mercury Panel Peer Review Letter: Review of EPA's Draft
National-Scale Mercury Risk Assessment
This appendix presents a copy of the SAB's Mercury Panel Peer Review Letter. The
original charge questions provided by EPA to the SAB are included at the end of the letter as
Appendix A.
120
-------
UNITED STATES ENVIRONMENTAL PROTECTION AGENCY
WASHINGTON D.C. 20460
OFFICE OF THE ADMINISTRATOR
SCIENCE ADVISORY BOARD
September 29, 2011
EPA-SAB-11-017
The Honorable Lisa P. Jackson
Administrator
U.S. Environmental Protection Agency
1200 Pennsylvania Avenue, N.W.
Washington, D.C. 20460
Subject: Review of EPA's Draft National-Scale Mercury Risk Assessment
Dear Administrator Jackson:
EPA's Office of Air and Radiation requested that the Science Advisory Board (SAB) review a draft
Technical Support Document: National-Scale Mercury Risk Assessment Supporting the Appropriate and
Necessary Finding for Coal and Oil-Fir ed Electric Generating Units -March 2011. The goal of this
draft document is to characterize human health exposure and risk associated with U.S. electrical
generating unit (EGU) mercury emissions with a focus on a highly exposed subpopulation, subsistence
fishers. The SAB was asked to comment on the risk assessment, including the overall design and
approach, as well as the use of specific models and key assumptions. The SAB was also asked to
comment on the extent to which specific facets of the assessment are well characterized in the Technical
Support Document.
The SAB could not evaluate the risk assessment based solely upon information provided in the
Technical Support Document. Important elements of the methods and findings are missing or poorly
explained. Additional information provided by EPA representatives during an SAB public meeting on
June 15-17, 2011 and a public teleconference on July 12, 2011 allowed the SAB to gain a sufficient
understanding of the risk assessment to conduct this review.
The risk assessment is designed to assess how a reduction in mercury emissions will translate to
reductions in fish tissue methylmercury concentrations, and in turn, to a reduction in potential risk to
subsistence fishers that would result from the consumption of self-caught fish from inland watersheds.
EPA sought advice from the SAB on key components of its analysis. In response, the SAB reviewed
available information and has made the following findings. The SAB considers the spatial resolution of
the modeling of mercury deposition to watersheds to be appropriate for the analysis. There is agreement
that the approach used to identify watersheds to include in the assessment is reasonable. This approach
is based upon the availability offish tissue methylmercury data and census data on target populations
with potential subsistence fishers. The SAB agrees that EPA's calculation of a hazard quotient for each
watershed is appropriate as the primary means of expressing risk because it is based on an established
reference dose for methylmercury that reflects a range of potential neurobehavioral effects. Intelligence
-------
Quotient (IQ) loss is also used in the assessment to evaluate risk. The SAB considers IQ loss to be an
insensitive indicator of methylmercury neurobehavioral effects and is concerned that its use as an
endpoint could underestimate risk. The SAB recommends that IQ loss be de-emphasized in the risk
assessment and explored as one of several possible secondary public health endpoints as a supplement to
the main analysis. Although the SAB considers the number of watersheds included in the assessment
adequate, some watersheds in areas with relatively high mercury deposition from U.S. EGUs were
under-sampled due to lack offish tissue methymercury data. The SAB encourages the Agency to contact
states with these watersheds to determine if additional fish tissue methylmercury data are available to
improve coverage of the assessment.
The SAB identifies additional sources of variability and uncertainty in the risk assessment, as well as
limitations imposed by the availability of data. The uncertainties are appropriate for a screening-level
public health assessment. The SAB regards the design of the risk assessment as suitable for its intended
purpose, to inform decision-making regarding an "appropriate and necessary finding" for regulation of
hazardous air pollutants from coal and oil-fired EGUs, provided that our recommendations are fully
considered in the revision of the assessment.
In summary, based on its review of the draft Technical Support Document and additional information
provided by EPA representatives during the public meetings, the SAB supports the overall design of and
approach to the risk assessment and finds that it should provide an objective, reasonable, and credible
determination of the potential for a public health hazard from mercury emitted from U.S. EGUs. The
SAB finds the current draft document deficient, however, because of its lack of transparency in
describing key analytical methods and findings. We urge the Agency to revise the document based on
our recommendations.
We appreciate the opportunity to review the draft mercury risk assessment. We look forward to your
response.
Sincerely,
/Signed/ /Signed/
Dr. Deborah L. Swackhamer Dr. Stephen M. Roberts
Chair Chair
Science Advisory Board SAB Mercury Review Panel
-------
NOTICE
This report has been written as part of the activities of the EPA Science Advisory Board (SAB), a public
advisory group providing extramural scientific information and advice to the Administrator and other
officials of the Environmental Protection Agency. The SAB is structured to provide balanced, expert
assessment of scientific matters related to problems facing the Agency. This report has not been
reviewed for approval by the Agency, and, hence, the contents of this report do not necessarily represent
the views and policies of the Environmental Protection Agency, nor of other agencies in the Executive
Branch of the Federal government. Mention of trade names of commercial products does not constitute a
recommendation for use. Reports of the SAB are posted on the EPA website at http://www.epa. gov/sab.
-------
U.S. Environmental Protection Agency
Science Advisory Board
Mercury Review Panel
CHAIR
Dr. Stephen M. Roberts, Professor, Department of Physiological Sciences, Director, Center for
Environmental and Human Toxicology, University of Florida, Gainesville, FL
MEMBERS
Dr. David T. Allen, Professor, Department of Chemical Engineering, University of Texas, Austin, TX
Dr. Thomas Burbacher, Professor of Environmental and Occupational Health Sciences, Director of the
Infant Primate Research Laboratory, Department of Environmental and Occupational Health Sciences,
School of Public Health, University of Washington , Seattle, WA
Dr. James Burch, Associate Professor, Department of Epidemiology and Statistics, Arnold School of
Public Health, University of South Carolina, Columbia, SC
Dr. Hillary Carpenter, Toxicologist, Health Risk Assessment, Environmental Health, Minnesota
Department of Health, St. Paul, MN
Dr. Celia Chen, Research Professor, Department of Biological Sciences, Dartmouth College, Hanover,
NH
Dr. Miriam L. Diamond, Professor, Department of Geography, University of Toronto, Toronto,
Ontario, CANADA
Dr. Charles T. Driscoll, Jr., Professor, Department of Civil and Environmental Engineering, College of
Engineering and Computer Science, Syracuse University, Syracuse, NY
Dr. Thomas M. Holsen, Professor, Department of Civil and Environmental Engineering, Clarkson
University, Potsdam, NY
Dr. James Hurley, Director, Environmental Health Division, Wisconsin State Laboratory of Hygiene,
and Associate Professor, Department of Civil and Environmental Engineering, University of Wisconsin-
Madison, Madison, WI
Dr. David Krabbenhoft, Research Scientist, Wisconsin Water Science Center, U.S. Geological Survey,
Middleton, WI
Dr. Leonard Levin, Technical Executive, Air Toxics Health & Risk Assessment, Environment Sector,
Electric Power Research Institute, Palo Alto, CA
Dr. C. Jerry Lin, Department of Civil Engineering, Lamar University, Beaumont, TX
11
-------
Dr. Jana Milford, Professor, Department of Mechanical Engineering, University of Colorado, Boulder,
CO
Dr. M. Christopher Newland, Alumni Professor, Department of Psychology, Auburn University,
Auburn, AL
Dr. Nicholas Ralston, Human Health Effects Research Group Leader, Energy & Environmental
Research Center (EERC), University of North Dakota , Grand Forks, ND
Dr. Stephen L. Rathbun, Professor of Biostatistics, Department of Epidemiology and Biostatistics,
University of Georgia, Athens, GA
Dr. Eric P. Smith, Professor, Department of Statistics, 406A Hutcheson Hall, Virginia Polytechnic
Institute and State University, Blacksburgh, VA
Dr. Alan Stern, Section Chief-Risk Assessment/ Adjunct Associate Professor, Division of Science,
Research & Technology/Dept. of Environmental & Occupational Health, New Jersey Department of
Environmental Protection/University of Medicine and Dentistry of New Jersey-Robert Wood Johnson
Medical School., Trenton, NJ (Affiliation for identification purposes only)
Dr. Edward Swain, Research Scientist, Minnesota Pollution Control Agency, Saint Paul, MN
Dr. Edwin van Wijngaarden, Associate Professor, Community and Preventive Medicine,
Environmental Medicine, and Dentistry, School of Medicine and Dentistry, University of Rochester,
Rochester, NY
Dr. Robert Wright, Associate Professor, Pediatrics, Division of Environmental Health, Harvard School
of Public Health, Boston, MA
SCIENCE ADVISORY BOARD STAFF
Dr. Angela Nugent, Designated Federal Officer, U.S. Environmental Protection Agency, Science
Advisory Board (1400R), 1200 Pennsylvania Avenue, NW, Washington, DC, Phone: 202-564-2218,
Fax: 202-565-2098, (nugent.angela@epa.gov)
in
-------
U.S. Environmental Protection Agency
Science Advisory Board
CHAIR
Dr. Deborah L. Swackhamer, Professor and Charles M. Denny, Jr., Chair in Science, Technology and
Public Policy, Hubert H. Humphrey School of Public Affairs and Co-Director of the Water Resources
Center, University of Minnesota, St. Paul, MN
SAB MEMBERS
Dr. David T. Allen, Professor, Department of Chemical Engineering, University of Texas, Austin, TX
Dr. Claudia Benitez-Nelson, Full Professor and Director of the Marine Science Program, Department
of Earth and Ocean Sciences , University of South Carolina, Columbia, SC
Dr. Timothy Buckley, Associate Professor and Chair, Division of Environmental Health Sciences,
College of Public Health, The Ohio State University, Columbus, OH
Dr. Patricia Buffler, Professor of Epidemiology and Dean Emerita, Department of Epidemiology,
School of Public Health, University of California, Berkeley, CA
Dr. Ingrid Burke, Director, Haub School and Ruckelshaus Institute of Environment and Natural
Resources, University of Wyoming, Laramie, WY
Dr. Thomas Burke, Professor, Department of Health Policy and Management, Johns Hopkins
Bloomberg School of Public Health, Johns Hopkins University, Baltimore, MD
Dr. Terry Daniel, Professor of Psychology and Natural Resources, Department of Psychology, School
of Natural Resources, University of Arizona, Tucson, AZ
Dr. George Daston, Victor Mills Society Research Fellow, Product Safety and Regulatory Affairs,
Procter & Gamble, Cincinnati, OH
Dr. Costel Denson, Managing Member, Costech Technologies, LLC, Newark, DE
Dr. Otto C. Doering III, Professor, Department of Agricultural Economics, Purdue University, W.
Lafayette, IN
Dr. David A. Dzombak, Walter J. Blenko Sr. Professor of Environmental Engineering , Department of
Civil and Environmental Engineering, College of Engineering, Carnegie Mellon University, Pittsburgh,
PA
Dr. T. Taylor Eighmy, Vice President for Research, Office of the Vice President for Research, Texas
Tech University, Lubbock, TX
Dr. Elaine Faustman, Professor, Department of Environmental and Occupational Health Sciences,
School of Public Health and Community Medicine, University of Washington, Seattle, WA
iv
-------
Dr. John P. Giesy, Professor and Canada Research Chair, Veterinary Biomedical Sciences and
Toxicology Centre, University of Saskatchewan, Saskatoon, Saskatchewan, Canada
Dr. Jeffrey Griffiths, Associate Professor, Department of Public Health and Community Medicine,
School of Medicine, Tufts University, Boston, MA
Dr. James K. Hammitt, Professor, Center for Risk Analysis, Harvard University, Boston, MA
Dr. Bernd Kahn, Professor Emeritus and Associate Director, Environmental Radiation Center, Georgia
Institute of Technology, Atlanta, GA
Dr. Agnes Kane, Professor and Chair, Department of Pathology and Laboratory Medicine, Brown
University, Providence, RI
Dr. Madhu Khanna, Professor, Department of Agricultural and Consumer Economics, University of
Illinois at Urbana-Champaign, Urbana, IL
Dr. Nancy K. Kim, Senior Executive, Health Research, Inc., Troy, NY
Dr. Kai Lee, Program Officer, Conservation and Science Program, David & Lucile Packard Foundation,
Los Altos, CA (affiliation listed for identification purposes only)
Dr. Cecil Lue-Hing, President, Cecil Lue-Hing & Assoc. Inc., Burr Ridge, IL
Dr. Floyd Malveaux, Executive Director, Merck Childhood Asthma Network, Inc., Washington, DC
Dr. Lee D. McMullen, Water Resources Practice Leader, Snyder & Associates, Inc., Ankeny, IA
Dr. Judith L. Meyer, Professor Emeritus, Odum School of Ecology, University of Georgia, Lopez
Island, WA
Dr. James R. Mihelcic, Professor, Civil and Environmental Engineering, State of Florida 21st Century
World Class Scholar, University of South Florida, Tampa, FL
Dr. Jana Milford, Professor, Department of Mechanical Engineering, University of Colorado, Boulder,
CO
Dr. Christine Moe, Eugene J. Gangarosa Professor, Hubert Department of Global Health, Rollins
School of Public Health, Emory University, Atlanta, GA
Dr. Horace Moo-Young, Dean and Professor, College of Engineering, Computer Science, and
Technology, California State University, Los Angeles, CA
Dr. Eileen Murphy, Grants Facilitator, Ernest Mario School of Pharmacy, Rutgers University,
Piscataway, NJ
Dr. Duncan Patten, Research Professor, Hydroecology Research Program , Department of Land
Resources and Environmental Sciences, Montana State University, Bozeman, MT
-------
Dr. Stephen Polasky, Fesler-Lampert Professor of Ecological/Environmental Economics, Department
of Applied Economics, University of Minnesota, St. Paul, MN
Dr. Arden Pope, Professor, Department of Economics, Brigham Young University, Provo, UT
Dr. Stephen M. Roberts, Professor, Department of Physiological Sciences, Director, Center for
Environmental and Human Toxicology, University of Florida, Gainesville, FL
Dr. Amanda Rodewald, Professor of Wildlife Ecology, School of Environment and Natural Resources,
The Ohio State University, Columbus, OH
Dr. Jonathan M. Samet, Professor and Flora L. Thornton Chair, Department of Preventive Medicine,
University of Southern California, Los Angeles, CA
Dr. James Sanders, Director and Professor, Skidaway Institute of Oceanography, Savannah, GA
Dr. Jerald Schnoor, Allen S. Henry Chair Professor, Department of Civil and Environmental
Engineering, Co-Director, Center for Global and Regional Environmental Research, University of Iowa,
Iowa City, IA
Dr. Kathleen Segerson, Philip E. Austin Professor of Economics , Department of Economics,
University of Connecticut, Storrs, CT
Dr. Herman Taylor, Director, Principal Investigator, Jackson Heart Study, University of Mississippi
Medical Center, Jackson, MS
Dr. Barton H. (Buzz) Thompson, Jr., Robert E. Paradise Professor of Natural Resources Law at the
Stanford Law School and Perry L. McCarty Director, Woods Institute for the Environment, Stanford
University, Stanford, CA
Dr. Paige Tolbert, Professor and Chair, Department of Environmental Health, Rollins School of Public
Health, Emory University, Atlanta, GA
Dr. John Vena, Professor and Department Head, Department of Epidemiology and Biostatistics,
College of Public Health, University of Georgia, Athens, GA
Dr. Thomas S. Wallsten, Professor and Chair, Department of Psychology, University of Maryland,
College Park, MD
Dr. Robert Watts, Professor of Mechanical Engineering Emeritus, Tulane University, Annapolis, MD
Dr. R. Thomas Zoeller, Professor, Department of Biology, University of Massachusetts, Amherst, MA
SCIENCE ADVISORY BOARD STAFF
Dr. Angela Nugent, Designated Federal Officer, U.S. Environmental Protection Agency, Science
Advisory Board (1400R), 1200 Pennsylvania Avenue, NW, Washington, DC, Phone: 202-564-2218,
Fax: 202-565-2098, (nugent.angela@epa.gov)
vi
-------
Table of Contents
Acronyms, Abbreviations and Definitions of Terms viii
1. Executive Summary 1
2. Introduction 6
3. Response to charge questions 7
3.1. Overall design 7
3.2. Critical health endpoints besides IQ loss 7
3.3. Use of an IQ loss metric benchmark 9
3.4. Spatial scale of watersheds 9
3.5. Measured fish tissue mercury concentrations 10
3.6. Use of the 75th percentile fish tissue methylmercury value 12
3.7. Consumption rates and location for high-end consumers 15
3.8. Use of Census data to identify high-end fish consuming populations 16
3.9. Use of the Mercury Maps approach 16
3.10. Exclusion of watersheds with significant non-air loadings 19
3.11. Concentration-response function used in modeling IQ loss 19
3.12. Uncertainty and variability 21
3.13. Discussion of analytical results 25
3.14. Responsiveness to the goals of the study 31
3.15. Confidence in the analysis 31
4. Summary list of recommendations 32
References 40
Appendix A: Agency Charge Questions A-l
vn
-------
ACRONYMS, ABBREVIATIONS AND DEFINITIONS OF TERMS
CASAC
CMAQ
EGU
EPA
GEOS-Chem
HAP
Hg
HQ
HUC
IQ
MDN
MeHg
MMAP
PUFA
RfD
R-MCM
SAB
TRI
TSD
Clean Air Scientific Advisory Committee
Community Multiscale Air Quality Modeling
System
Electrical Generating Unit
Environmental Protection Agency
A global 3-D chemical transport model (CTM) for
atmospheric composition driven by meteorological
input from the Goddard Earth Observing System
(GEOS) of the NASA Global Modeling and
Assimilation Office.
Hazardous Air Pollutant
Mercury
Hazard Quotient
Hydrologic Unit Codes
Intelligence Quotient
Mercury Deposition Network
Methylmercury
Mercury Maps
Polyunsaturated Fatty Acid
Reference Dose
Regional Mercury Cycling Model
Science Advisory Board
Toxic Release Inventory
Technical Support Document
Vlll
-------
1. Executive Summary
EPA has proposed National Emission Standards for Hazardous Air Pollutants for coal- and oil-fired
Electric Utility Steam Generating Units (EGUs). These proposed standards would require EGUs to
decrease emissions of mercury and other hazardous air pollutants (HAP). In order to regulate HAP
emissions under the Clean Air Act, Section 112(b), the Agency must make a determination that such
regulation is appropriate and necessary based upon a study of the hazards to public health reasonably
anticipated from HAP emissions. As part of this determination, hazards to public health from U.S. EGU
mercury emissions are evaluated in a draft national-scale risk assessment entitled Technical Support
Document: National-Scale Mercury Risk Assessment Supporting the Appropriate and Necessary Finding
for Coal and Oil-FiredElectric Generating Units (March 2011). This SAB report uses the terms "risk
assessment" and "Technical Support Document" interchangeably to refer to EPA's draft document.
The draft risk assessment considers hazards from mercury released from U.S. EGUs and deposited in
watersheds within the continental United States. Mercury deposition is estimated using the Community
Multi-scale Air Quality (CMAQ) model for watersheds classified using 12-digit Hydrologic Unit Codes
(HUC12). The risk assessment focuses on hazard from consumption of methylmercury in self-caught
fish, specifically hazard to children born to women who consume local fresh water fish in a subsistence
manner. Exposure from fish consumption is estimated for watersheds with data on methylmercury
concentrations in fish tissue, and a hazard quotient (HQ) is calculated based upon the current reference
dose (RfD) for methylmercury. The contribution of U.S. EGUs to the HQ for each watershed is
calculated by comparing U.S. EGU deposition rates with total deposition to the watershed, including
other sources, assuming that the contribution of U.S. EGUs to fish tissue concentrations and risk is
proportional to their contribution to total deposition. Intelligence Quotient (IQ) loss is also modeled as a
health endpoint, with a loss of one or more points from methylmercury exposure considered as a public
health concern. Estimated hazards associated with U.S. EGU emissions in 2005 are compared with
estimated hazards expected to remain in 2016 "after imposition of the requirements of the Act."
The SAB Mercury Review Panel was asked to comment on the draft risk assessment, including the
overall design and approach as well as various technical aspects. The Panel was also asked to comment
on the extent to which specific "observations" or conclusions in the risk assessment are supported by the
analytical results. During the course of deliberations, the Panel reviewed background materials provided
by the Office of Air Quality Planning and Standards, as well as public comments on the topic. The SAB
reviewed and approved the report of the Panel. EPA asked the SAB to address fourteen charge
questions, many with multiple parts. This Executive Summary highlights the main findings. Detailed
responses to the individual charge questions are provided in the body of the report.
The SAB finds the Technical Support Document to lack critical details regarding both the methods used
and the results presented. This made the document difficult to review and, in the view of the SAB,
unsuitable in its present form to fully support Agency decision-making. Presentations and information
provided by Agency representatives helped the SAB understand how the risk assessment was conducted,
the rationale for some of the decisions made in approach and the use of data and the translation of the
results. With this additional information, the SAB views the risk assessment favorably, concluding that
it is able to provide an objective, reasonable, and credible determination of the potential for a public
health hazard from mercury emitted from U.S. EGUs. However, the SAB considers the integrity of the
risk assessment to be dependent in part on a transparent description of the analysis, and the Technical
Support Document needs to be strengthened to provide this description.
1
-------
This review is based on the text of the report and the additional information provided by the above cited
presentations and discussion provided by EPA representatives. Responses to charge questions indicate
where improvements need to be made, and a summary of the most critical recommendations is provided
in section 4. The SAB's support for the risk assessment is contingent on its recommendations being fully
considered in the revision of the assessment.
Overall design
The Panel finds the overall design and general approach used in the risk assessment to be scientifically
credible. The Technical Support document, however, needs a more detailed description of the modeling
methods and data sources. The report's introduction should make clear from the start that the analysis is
a determination of potential exposure at the scale of watersheds.
Critical health endpoints besides IQ loss; use of an IQ loss metric benchmark; and concentration-
response function used in modeling IQ loss
The SAB supports the use of the HQ approach in the risk assessment. SAB members agree that because
the RfD from which the HQ is calculated is an integrative metric of neurodevelopmental effects of
methylmercury, it constitutes a reasonable basis for assessing risk. Other potential health endpoints were
also considered by the SAB. The SAB notes that a number of measures of potential neurodevelopmental
effects of methylmercury exist, some of which have greater sensitivity to differential mercury exposure
than does IQ loss. However, none are viewed by the SAB as suitable for quantitative risk estimation
with a reasonable degree of scientific certainty at the present time, and consequently none are
recommended for incorporation into the analysis. The SAB does not consider it appropriate for EPA to
use IQ loss in the risk assessment and recommends that this aspect of the analysis be de-emphasized,
moving it to an appendix where IQ loss is discussed along with other possible endpoints not included in
the primary assessment.
While the SAB agrees that the concentration-response function for IQ loss used in the risk assessment
has validity, IQ loss is not a sensitive response endpoint for methylmercury and its use likely
underestimates the impact of reducing methylmercury in water bodies. The SAB agrees that if IQ loss
were retained in the risk assessment despite these reservations, a loss of one or two points on average in
a population would be an appropriate benchmark. The SAB agrees that fish nutrients (e.g., omega-3
fatty acids) can potentially ameliorate neurologic effects associated with methylmercury, but there is not
sufficient information to recommend a quantitative adjustment in health endpoint measures. However,
the SAB agrees that because the RfD from which the HQ is calculated is an integrative metric of risk, it
constitutes a reasonable basis for assessing risk. The RfD is an integrative measure because it considers
the weight of the evidence and determines a quantitative value that is based on the most sensitive
endpoints across multiple studies and endpoints.
Spatial scale of watersheds
The SAB agrees that HUC12 watersheds provide the appropriate level of spatial resolution and offer
advantages over previous assessments at lower resolution (e.g., HUC8). The comparability of this scale
to CMAQ output makes the transferability and applicability of deposition modeling to the watershed
scientifically robust. Further, the finer resolution of HUC12 watersheds is better suited to follow
deposition patterns of a single source such as an EGU and increases the likelihood that measured
deposition within a watershed is homogeneous. The SAB notes that one disadvantage of smaller
-------
watershed size is that, within a given watershed, the number offish samples with methylmercury data is
diminished. During the public meeting, the SAB questioned some of the figures with maps showing
modeled deposition across the United States. Some areas showed intense deposition with no obvious
source, leading SAB members to question the accuracy of the modeling or data presentation in the
March draft report. EPA provided clarification and updated maps in July 2011 (Pekar 2011). The SAB
supports EPA's plans to include these updated mercury deposition maps in the revised report so they
correctly reflect total annual mercury deposition per square meter by watershed.
Measured fish tissue mercury concentrations
The SAB agrees that fish tissue methylmercury data are an appropriate basis to estimate the number and
percentage of fish-sampled watersheds where populations may be at risk. Although fish data were only
available for 2,461 HUC12 watersheds out of 88,000 HUC12 watersheds in the continental United
States, this is viewed as sufficient to estimate the number and percentage offish-sampled watersheds
where populations may be at risk. The SAB notes advantages and disadvantages of the Agency decision
to limit fish tissue concentration data to the period after 1999 but agrees with this approach, given that
older data might not be representative of conditions during the 2005 reference deposition year. The SAB
is concerned about the absence offish tissue data from some watersheds with higher levels of mercury
deposition. The EPA is encouraged to contact states with these watersheds to determine whether
additional fish tissue data are available to improve coverage of the analysis. The SAB discussed the use
of modeling to estimate fish methylmercury concentrations as a means to include more watersheds. With
further development, this approach could be used for a national scale assessment such as this in the
future but the SAB does not recommend it for the current assessment.
Use of the 75th percentile fish tissue methylmercury value
As a means of selecting methymercury fish concentrations representative of larger, but not the largest,
edible fish, the risk assessment uses the 75th percentile fish concentration for watersheds with one or
more fish concentration value. The SAB considers this percentile reasonable but is concerned that over
half of the watersheds in the assessment have only four or fewer fish samples with methylmercury
concentration, and a significant number of these have a single fish sample. The SAB notes that in
watersheds where only a few fish samples are available, the 75th percentile concentration and exposure
most likely will be underestimated. This should be explained in the report, and a sensitivity analysis
should be conducted using the median fish tissue concentration to better represent the distribution of
concentrations when the sample size is only one fish. The SAB also recommends that the report describe
the sources offish methylmercury concentration data more fully, including the state sampling programs
that provide most of the data. Discussion of sampling programs should include the types and sizes of
fish obtained, as well as uncertainties associated with this data set, to improve the transparency of the
analysis.
Consumption rates and location for high-end consumers
The SAB finds that the consumption rates and locations for fishing activity for likely highly exposed
consumers, i.e., self-caught fish consuming populations modeled in the analysis, are supported by the
data presented in the document and are generally reasonable and appropriate given the available data. A
diverse range of susceptible populations is represented in the assessment. There are caveats, however,
associated with the sources offish consumption data, the data sets selected for inclusion, and the
-------
suitability of data for inclusion in the risk assessment (e.g., in terms of providing annual average intakes
of the edible portion of the fish) that should be acknowledged more fully in the document.
Use of Census data to identify high-end fish consuming populations
The SAB agrees that the criterion of using at least 25 persons per census tract from a given target
subsistence fisher population is a reasonable approach to identify watersheds with potentially highly
exposed fish consuming populations. While other approaches are possible, none is viewed as being more
effective or feasible. The SAB recommends that the document clarify how many watersheds were
eliminated due to this inclusion criterion.
Mercury Maps approach
The SAB agrees with the Mercury Maps approach used in the analysis and has cited additional work that
supports a linear relationship between mercury loading and accumulation in aquatic biota. These studies
suggest that mercury deposited directly to aquatic ecosystems can become quickly available to biota and
accumulated in fish, and reductions in atmospheric mercury deposition should lead to decreases in
methylmercury concentrations in biota. The SAB notes other modeling tools are available to link
deposition to fish concentrations, but does not consider them to be superior for this analysis or
recommend their use. The integration of Community Multiscale Air Quality Modeling System (CMAQ)
deposition modeling to produce estimates of changes in fish tissue concentrations is considered to be
sound. Although the SAB is generally satisfied with the presentation of uncertainties and limitations
associated with the application of the Mercury Maps approach in qualitative terms, it recommends that
the document include quantitative estimates of uncertainty available in the existing literature.
Exclusion of watersheds with significant non-air loadings
In order to reduce uncertainty associated with the Mercury Maps approach, watersheds with significant
non-air loadings of mercury are excluded from the analysis. The SAB agrees with the exclusion criteria
used by the Agency. Additional exclusion criteria were discussed, but their application would be
unlikely to substantially change the results of the assessment. The SAB also recommends that the EPA
provide additional discussion of uncertainties in the mercury emissions from U.S. EGUs and non-EGU
sources and the implications of these uncertainties.
Uncertainty and variability and discussion of analytical results
The SAB discussed the characterization of variability and uncertainty in the Technical Support
Document in detail. Sources of variability and uncertainty in the assessment are summarized in
Appendix F of the draft document. The qualitative nature of this presentation is considered appropriate,
but the identification of important sources of variability and uncertainty is considered incomplete.
Inclusion of several additional sources of variability and uncertainty is recommended. The SAB notes
that the degree of uncertainty associated with the analysis is consistent with a screening level analysis,
and despite the various sources of uncertainty inherent in the approach, the analysis is sound and
reasonable.
The SAB finds that observations in five areas (mercury deposition from U.S. EGUs, fish tissue
methylmercury concentrations, patterns of mercury deposition with mercury fish tissue data, percentile
risk estimates, and number and frequency of watersheds with populations potentially at risk due to U.S.
-------
EGU mercury emissions) are generally supported by the analytical results presented in the document.
However, there are many examples where results are poorly presented, and in most areas the
uncertainties, variability and data limitations are not well characterized. The SAB has numerous specific
recommendations to improve presentation of findings and observations.
Responsiveness to the goals of the study
The section of the document on Summary of Key Observations does not encapsulate well the critical
issues and significant results of the analysis. The SAB recommends revising this section to link back
directly with the goals of the studies as articulated on Page 13 of the document, i.e.: (a) what is the
nature and magnitude of the potential risk to public health posed by current U.S. EGU mercury
emissions? (b) what is the nature and magnitude of the potential risk posed by U.S. EGU mercury
emissions in 2016 considering potential reductions in EGU Hg emissions attributable to CAA (Clean Air
Act) requirements? and (c) how is risk estimated for both the current and future scenario apportioned
between the incremental contribution from U.S. EGUs and other sources of mercury?
-------
2. Introduction
EPA's Office of Air and Radiation requested peer review of a Technical Support Document: National-
Scale Mercury Risk Assessment Supporting the Appropriate and Necessary Finding for Coal and Oil-
FiredElectric Generating Units -March 2011, developed to support a proposed rule published in the
Federal Register on March 16, 2011 to regulate emissions of hazardous air pollutants from for coal- and
oil-fired Electric Utility Steam Generating Units (EGUs). Section 112(n)(l) of the Clean Air Act
requires EPA to determine whether it is "appropriate and necessary" to regulate hazardous air pollutants
emissions from EGUs under section 112. The "appropriate and necessary" finding requires EPA to
perform a study of the hazards to public health reasonably anticipated to occur as a result of hazardous
air pollutant emissions, including mercury.
The Science Advisory Board formed an expert ad hoc Panel to peer review the draft Technical
Document. The Panel addressed fourteen Agency charge questions (see Appendix A) and developed the
responses below. The Panel held a public meeting on June 15-17, 2011 to peer review this document and
held a public teleconference on July 20, 2011 to discuss the Panel's draft report. The chartered SAB held
a quality review to approve the draft report on September 7, 2011.
The SAB had difficulty evaluating the Technical Support Document because it lacked critical details.
During the public meeting, presentations and information provided by Agency representatives helped
the SAB understand technical aspects of the analysis. With this additional information and clarification,
the SAB views the risk assessment positively. However, the SAB considers the integrity of the risk
assessment as dependent in part on a transparent description of the methods and findings. The Technical
Support Document needs to better explain what was done and why, translate the results into findings
that relate to the key goals of the analysis and describe where the uncertainties lie. The SAB's support
for the risk assessment is contingent upon a development of a revised document that addresses these
issues.
The body of this report is organized to respond to each of EPA's charge questions. Section 3 provides
responses and specific suggestions and recommendations for revising the Technical Support Document.
Because this SAB report provides many recommendations for strengthening the Technical Support
Document, Section 4 provides a list of the specific recommendations made by the SAB.
-------
3. Response to Charge Questions
3.1. Overall design
Question 1: Please comment on the scientific credibility of the overall design of the mercury risk
assessment as an approach to characterize human health exposure and risk associated with U.S.
EGU mercury emissions (with a focus on those more highly exposed).
Response: The SAB finds that the overall design and general approach used in the assessment are
scientifically credible.
The overall approach used in the study is to estimate potential risk at a national scale, attributable to
mercury released from U.S. EGUs and deposited to inland waterbodies, for recent (2005) and future
(2016) emissions levels. To accomplish this, the analysis links a series of models and data to estimate
methylmercury exposure via fish consumption and then compares the exposure with a toxicological
benchmark. The series of models allows for the estimation of deposition of mercury emitted by U.S.
EGUs into watersheds. The assessment uses measured concentrations of methylmercury in fish tissue
samples, as well as estimates of future fish methylmercury concentrations, to estimate the number and
percentage of watersheds where populations may be at risk. Human exposure and potential health effects
in these at-risk watersheds are then assessed through the pathway of ingestion of self-caught fish from
inland water bodies for vulnerable subsistence fisher populations.
Although the overall design and general approach are scientifically credible, the SAB has a number of
suggestions and recommendations for enhancing the assessment, based on review of the draft Technical
Support Document and supplementary presentations and information provided by EPA. The responses to
the charge questions below provide those recommendations and suggestions in detail. It will be
important for EPA to address these issues. The Technical Support Document would benefit from a more
detailed description of the modeling methods and data sources, and results need to be presented more
clearly. The Introductory section should make clear, at the earliest possible point, that the analysis is a
determination of watershed impact with exposure addressed as a potential outcome. Despite weaknesses
in the Technical Support Document and uncertainties inherent in an analysis such as this, the SAB
agrees that the risk assessment makes an objective and reasonable determination of the potential for a
public health hazard from mercury emitted from U.S. EGUs.
3.2. Critical health endpoints besides IQ loss
Question 2: Are there any additional critical health endpoint(s) besides IQ loss which could be
quantitatively estimated with a reasonable degree of confidence to supplement the mercury risk
assessment (see section 1.2 of the Mercury Risk TSDfor an overview of the risk metrics used in the risk
assessment)?
Response: This charge question raises issues about the use of IQ, as well as use of alternative
quantitative measures. While several alternative approaches were discussed that might supplement IQ
scores, no substitute can be quantitatively estimated with a "reasonable degree of confidence."
Moreover, there are doubts that IQ met this standard. The response to this charge question addresses
both of these issues below.
-------
Use oflQ. There are significant concerns about the use of IQ for identifying the impact of consuming
fish from water bodies with unacceptable levels of methylmercury because IQ will likely result in an
underestimation of potential neurobehavioral impacts, compared to analyses using the hazard quotient
(HQ). Thus, the SAB considers HQ to be a stronger basis for evaluation of methylmercury hazard. The
HQ is based upon the methylmercury reference dose (RfD), which is an integrative measure reflecting a
range of neurobehavioral effects, and it incorporates pharmacokinetic variability. The RfD considers the
weight of the evidence and determines a quantitative value that is based on the most sensitive endpoints
across multiple studies and endpoints. Sensitive endpoints, in this context, are adverse effects that occur
at the lowest exposures.
In contrast, the loss of IQ points is likely to underestimate the impact of reducing methyl mercury in
water bodies. The reason is that IQ score has not been the most sensitive indicator of methylmercury's
neurotoxicity in the populations studied. As noted in the Technical Support Document, in the Faroe
Island study the most sensitive indicators were in the domains of language (Boston Naming Test),
attention (continuous performance) and memory (California Verbal Learning Test). These two tests are
neuropsychological tests that are not subtests of IQ tests and whose relationship with global IQ is not
well characterized. In the Seychelles study, the Psychomotor Development Index was the most sensitive
measure and, while this index is a component of the Bailey Scales of Infant Development, it is not
highly correlated with cognitive measures (Davidson et al. 2008).
Additionally, the use of IQ, or any neuropsychological measure, distracts from the main goal of the
document. The analysis in the document emphasizes the number offish-sampled water bodies from
which subsistence fishers would be at risk based on an elevated HQ. As is clear in Tables 2-9 to 2-11 in
the Technical Support Document, an analysis based on IQ identifies far fewer water bodies than one
based on the HQ. This is because IQ underestimates hazard, as noted above.
The SAB recommends that EPA reframe the document's discussion of IQ. EPA should incorporate IQ
and other neuropsychological measures as supplemental information and focus on HQ as the primary
critical health endpoint. It is not suggested that the analyses of IQ be removed altogether but rather that
the analyses be framed in an appendix to the report as a secondary analyses of impact of reduced
exposure on potential health-related outcome. The appendix should discuss the basis for selecting a HQ
at or above 1.5 as the criteria for selecting potentially impacted watersheds should be explained. The
appendix should also include discussion of potential effects on other measures like developmental
delays (Grandjean et al. 1997) or neuropsychological tests (as discussed by van Wijngaarden et al.
2006), presented in the overall context of the weight of evidence.
Alternative quantitative measures. One alternative is developmental delay as described by Grandjean et
al. (1997). Here, an estimate of the number of months of delay in verbal skills as tapped by the Boston
Naming Test or in learning and short-term memory as tapped by the California Verbal Learning Test
was made based on regression coefficients describing the relationship among age, methylmercury
exposure, and scores on these tests. The delays were on the order of five to seven months associated
with a 10-fold increase in cord blood methylmercury.
A recent analysis by van Wijngaarden et al. (2006) derived Benchmark Dose Level-Lower 95%
confidence interval values for 26 endpoints, including IQ and other neuropsychological measures from
the nine-year follow up of the Seychelles child development study main cohort. This paper could be
cited in a discussion of markers of health impacts of lowering mercury deposition and reducing intake
by subsistence fishers.
-------
One SAB member suggests the use of blood markers of selenium-dependent enzyme function, noting
that methylmercury irreversibly inhibits selenium-dependent enzymes that are required to support vital-
but-vulnerable metabolic pathways in the brain and endocrine system. Impaired selenoenzyme activities
would be observed in the blood before they would be observed in brain, but the effect is also expected to
be transitory. The use of these measures is a minority view among the SAB members.
The SAB recommends that the Technical Support Document acknowledge and discuss alternative
quantitative measures but does not recommend a re-analysis based on these measures.
3.3. Use of an IQ loss metric benchmark
Question 3: Please comment on the benchmark used for identifying a potentially significant public
health impact in the context of interpreting the IQ loss risk metric (i.e., an IQ loss of 1 to 2 points or
more representing a potential public health hazard). Is there any scientifically credible alternate
decrement in IQ that should be considered as a benchmark to guide interpretation of the IQ risk
estimates (see section 1.2 of the Mercury Risk TSDfor additional detail on the benchmark used for
interpreting the IQ loss estimates).
Response: The consensus is that if IQ were to be used, then a loss of 1 or 2 points as a population
average is a credible decrement to use for this risk assessment. This metric seems to be derived from the
lead literature and was peer-reviewed by the Clean Air Scientific Advisory Committee (U.S. EPA
CAS AC 2007). While its applicability to methylmercury is questionable, the size of the decrement is
justified based on the extensive analyses available from the literature reviewed by CASAC. The support
for the model of the relationship between IQ and methylmercury exposure comes from Axelrad and
Bellinger (2007) and from a whitepaper produced by Bellinger (2005).
The analysis in Table 2-10 showing the effect of using a one- or two -point loss is helpful in evaluating
the sensitivity of this measure to the magnitude of the decrement.
3.4. Spatial scale of watersheds
Question 4: Please comment on the spatial scale used in defining watersheds that formed the basis for
risk estimates generated for the analysis (i.e., use of 12-digit hydrologic unit code classification). To
what extent do HUC12 watersheds capture the appropriate level of spatial resolution in the relationship
between changes in mercury deposition and changes inMeHgfish tissue levels? (see section 1.3 and
Appendix A of the Mercury Risk TSDfor additional detail on specifying the spatial scale of watersheds
used in the analysis).
Response: The choice of using the HUC12 (Hydrologic Unit Code) watershed delineation of the
contiguous 48 United States for this risk assessment is more appropriate and offers at least two distinct
advantages over the 2001 Mercury Maps study report that employed the larger-scale HUC8 delineation.
First, HUCSs are "cataloguing units" delineation and do not actually represent true watersheds (areas of
land where surface water drainage accumulates to an outflow location). Instead, many HUC8 areas have
flow lines that cross the unit boundaries, thus making this larger scale delineation not technically correct
for any mass accounting procedure like Mercury Maps. The use of HUC12s, which are true watershed
delineations, does not violate this mass accounting assumption. A second strength of the use of
HUC12's is that they have a similar physical scale to the spatial resolution of the CMAQ output (12 km
CMAQ square grid compared to the HUC12 watersheds that are typically about 5-10 km on a side).
Comparable scales make the transferability and applicability of deposition modeling to the watershed
-------
more scientifically robust. The use of finer scale watersheds enables modeling and deposition runs that
have the detail to follow deposition patterns from a single source, including EGUs. The fine-scale
watershed resolution decreases the likelihood that there is a significant deposition gradient within the
HUC. Further, the relative biogeochemical and ecological homogeneity of an individual HUC12
watershed allows better validity for ascribing fish concentrations to a specific watershed and that those
fish will respond in proportion to changes in atmospheric mercury deposition. The SAB notes, however,
that one potential disadvantage of HUC12 is that a number of HUC12 watersheds contain a very limited
number offish samples because of their inherent small size, but other factors described in this response
override this concern.
The Technical Support document acknowledges and this SAB agrees that the fish distribution data are
highly skewed toward the Eastern United States. That said, the legend of Figure 2-6 in the Technical
Support Document indicates that 2,170 out of 2,461 watersheds were from the Eastern United States,
leaving approximately 300 samples from Western sites. Given the apparent distribution of high
deposition zones in CMAQ modeling runs displayed in Figures 2-1 and 2-2 in the Technical Support
Document that are not ground-truthed in Mercury Deposition Network deposition measurement, the
SAB is concerned not only about the reality of the identified intense deposition zones (i.e., whether they
are truly intense deposition zones, for example, in the state of Nevada), but also whether these
watersheds were included in this report's analysis. Fish distribution data appear to overlap with some of
these zones of modeled high mercury deposition, and, with 300 fish samples from the Western United
States, there is a high probability for overlap.
The SAB is concerned about the possibility that in some watersheds, multiple small lakes may be
included within a single HUC12. In some cases, lakes within a small geographic zone have been shown
to have quite different chemistry and biological productivity. For instance, within Voyageurs National
Park in northern Minnesota, the mercury content of similarly-sized fish of a given species in about 20
lakes varies by a factor of 10 (Wiener et al. 2006), indicating that even lakes near each other can
bioaccumulate mercury to greatly differing degrees. In HUCs with multiple lakes, the SAB recommends
against using a single fish methylmercury value to describe the HUC. In response to this concern and
other charge questions, the SAB recommends that the authors provide a summary table describing the
characteristics of the watersheds where fish were collected, including the fraction offish samples
collected from rivers versus lakes, and whether from single or multiple sites.
3.5. Measured fish tissue mercury concentrations
Question 5: Please comment on the extent to which the fish tissue data used as the basis for the risk
assessment are appropriate and sufficient given the goals of the analysis. Please comment on the extent
to which focusing on data from the period after 1999 increases confidence that the fish tissue data used
are more likely to reflect more contemporaneous patterns of mercury deposition and less likely to reflect
earlier patterns of mercury deposition. Are there any additional sources offish tissue MeHg data that
would be appropriate for inclusion in the risk assessment?
Response: The measured fish tissue data serve as an appropriate basis for the mercury risk assessment
because they are widely available and reflect the actual environmental conditions that influence fish
methylmercury concentrations and human exposure to methylmercury by the target populations. The
SAB notes that the relevant form of mercury in fish tissue for this risk assessment is methylmercury, but
there is sometimes ambiguity as to the mercury form actually measured in surveys from which the fish
tissue data were taken. Many surveys measure total mercury and assume all mercury present in fish is in
10
-------
the methyl form. Although empirical data available are largely supportive of this assumption, the
Technical Support Document needs to clearly acknowledge this aspect of the fish tissue data.
While it is always desirable to have a larger sample size, the sample size of 2,461 HUC12 watersheds is
adequate for the goals of the risk assessment. However, as detailed below, the SAB is concerned about
the sources of bias and uncertainty resulting from the state sampling designs used to select watersheds
where fish tissue samples were obtained. For purposes of hazard assessment, it is reasonable to have an
over-representation of HUC12s in the eastern part of the country given the prevalence of EGUs in the
East. However, the description of the character of the data, as well as the selection of analyzable data
(e.g., sizes, distribution offish sizes across watersheds), should be better detailed in the report.
There are advantages and disadvantages to using fish methylmercury data prior to 1999 for the risk
assessment. The advantage is that considerable fish data were obtained prior to 1999 and the use of these
data could increase the information available for the national risk assessment. The disadvantage is that
fish methylmercury concentrations may have changed since 1999 and these older data may not be
representative of conditions during the 2005 reference deposition year. Unfortunately, there are few high
quality time series data offish methylmercury concentrations, so it is difficult to quantify the extent to
which fish methylmercury concentrations have changed since the 1990s. As a result, the SAB
recommends that the EPA utilize fish methylmercury data collected since 1999 for the risk assessment.
Given the spatial distribution of mercury deposition from EGUs and the density offish methylmercury
measurements (Figure 2-15), there are some states that receive what the Technical Support Document
terms "relatively elevated" mercury deposition from U.S. EGU emissions and have limited fish
methylmercury measurements. These states include Pennsylvania, New Jersey, Kentucky and Illinois.
The SAB suggests that the EPA contact these states to investigate if additional recent (since 1999) fish
methylmercury data are available to improve the coverage for the mercury risk assessment. For example,
the Pennsylvania Department of Environmental Protection, Pennsylvania Fish Monitoring Program has
700 sites for the measurement of the methylmercury content of recreational sport fish, with samples
collected from 1979-2007.
EPA's reliance on the National Listing of Fish Advisory and U.S. Geological Survey (USGS)
compilation of methylmercury data sets contributes to uncertainty because these data were collected by
state agencies with various sampling designs and state protocols. Most of the data are not from
probability-based sampling designs, so it is not entirely clear what population the fish tissue samples
represent. The direction of impact on the risk assessment of this variation in sampling designs cannot be
ascertained. Moreover, some states have greater sampling efforts than others. Particularly strong
sampling efforts were observed in South Carolina, Louisiana, Indiana, Iowa, West Virginia and Virginia.
As a consequence of this variability in fish-tissue sampling effort, the risk assessment will be strongly
influenced by states with high sampling efforts. Moreover, Figure 2-18 suggests that the sample is
biased in favor of watersheds with higher mercury deposition and higher EGU-attributable deposition as
predicted by the CMAQ model. This bias could in part be due to the over-representation of HUC12s in
the East but could also occur if states with high deposition also have high fish-tissue sampling effort.
Nevertheless, as per the limitations of the available data, the risk assessment focuses on that portion of
the fish-sampled watersheds at risk, rather than attempting to make inferences to the larger population of
all 88,000 HUC12 watersheds.
Researchers have developed empirical models for fish methylmercury concentrations using water
chemistry and land cover data (Chen et al. 2005; Driscoll et al. 2007; and Watras et al. 1998). These
11
-------
empirical relationships have been used to estimate methylmercury concentrations for different fish
species at state and regional spatial scales. Such an empirical modeling approach could be used to
provide more comprehensive estimates offish methylmercury concentrations across water resources and
potentially improve the extent of future mercury risk assessments. However, if this empirical modeling
approach was to be used in a risk assessment such as this, it would need to be developed and evaluated
at a national scale. Moreover, empirical models would contribute additional uncertainty in the estimation
offish methylmercury concentration. The SAB is not recommending that this approach be used for the
current risk assessment. Rather, the EPA might consider use of empirical modeling to improve the
information available related to fish methylmercury concentrations in future assessments.
To strengthen the Technical Support Document the SAB recommends that it be revised to provide a
better description of the character of the data, as well as the selection of analyzable data (e.g., sizes,
distribution offish sizes across watersheds). The SAB also recommends that EPA contact some states
that receive what the Technical Support Document terms "relatively elevated" mercury deposition from
U.S. EGU emissions and have limited fish methylmercury measurements to investigate if additional
recent (since 1999) fish methylmercury data are available to improve the coverage for the mercury risk
assessment.
3.6. Use of the 75th percentile fish tissue methylmercury value
Question 6: Given the stated goal of estimating potential risks to highly exposed populations, please
comment on the use of the 75th percentile fish tissue MeHg value (reflecting targeting of larger but not
the largest fish for subsistence consumption) as the basis for estimating risk at each watershed. Are
there scientifically credible alternatives to use of the 75th percentile in representing potential
population exposures at the watershed level?
Response: Using the 75th percentile offish tissue values as a reflection of consumption of larger, but not
the largest, fish among sport and subsistence fishers is a reasonable approach and is consistent with
published and unpublished data on predominant types offish consumed. While the choice of the 75th
percentile is reasonable for the estimation of the methymercury levels of consumed fish, the
appropriateness of this approach depends on the data from which the value is derived. The SAB is
concerned that around 29% of watersheds sampled have only one fish sample with a fish tissue
methylmercury concentration available. Figure 1 below shows a plot of the number offish tissue
samples available for rivers (N= 1551 samples from rivers, 41.5% have one fish measurement) using data
provided to the SAB by EPA. There is clear evidence of a very high proportion of samples with only one
fish.
12
-------
600-
400-
cr
a
PL,
200-
n
nn—in—ii—ir-
I I I I I I I I I I I I I I I I I I I I I I I I I
v>
Number of samples per watershed
Figure 1. Frequency of fish samples of different sizes for rivers using Excel data provided to the SAB. The \ axis
corresponds to the number of fish tissue observations per HUC. When sample sizes are 20 or greater, a category is
used i.e. 20 corresponds to 20 to 25,25 corresponds to 26 to 30, etc.
Thus, the estimate of the 75th percentile has considerable uncertainty. The use of only one tissue value
for a given watershed is likely to underestimate fish tissue levels if the single fish collected was, on
average, smaller than the true 75th percentile, as would occur if the collection were random. Support for
this notion is provided by Figure 2 below, which relates the 75th percentile fish tissue methylmercury
concentration (on y axis) to the number offish samples available for any given watershed. The estimate
of the 75* percentile appears to increase with increasing sample size, thus suggesting that the 75*
percentile fish tissue concentration for watersheds with few fish samples is underestimated.
13
-------
S 2.0-
1
u
§
o
1.5-
1.0-
•a
is
o
vg 0.5 -
o.o H
o
o cf
-------
the size offish sampled in watersheds with few fish tissue samples available on estimated mercury
concentrations. The SAB also recommends that the Technical Support Document clarify that the 75*
percentile represents available fish tissue data that may or may not represent the fish in the watershed or
the fish consumed.
3.7. Consumption rates and location for high-end consumers
Question 7: Please comment on the extent to which characterization of consumption rates and the
potential location for fishing activity for high-end self-caught fish consuming populations modeled in the
analysis are supported by the available study data cited in the Mercury Risk TSD. In addition, please
comment on the extent to which consumption rates documented in Section 1.3 and in Appendix C of the
Mercury Risk TSD provide appropriate representation of high-end fish consumption by the subsistence
population scenarios used in modeling exposures and risk. Are there additional data on consumption
behavior in subsistence populations active at inland freshwater water bodies within the continental
U.S.?
Response: The SAB finds that the consumption rates and locations for fishing activity are supported by
data presented in Section 1.3 and in Appendix C of the Technical Support Document. In addition, the
targeted locations and fish consumption data used in the analysis are generally appropriate and
reasonable given the available data. The risk assessment uses sources that reported daily consumption
for populations of low socioeconomic status African- and European-Americans females as the target
population for the risk assessment. In addition, consumption rates from a study that targeted Laotian-
and Vietnamese-Americans, previously identified in the central valley of California, are included in the
assessment, as well as those from a study of Great Lakes tribes. Thus, a diverse range of susceptible
populations is represented in the assessment.
The SAB recommends that a few caveats should be acknowledged more fully in the document. The
main consumption estimates comes from a relatively small survey of individuals attending a fishing
convention in South Carolina, so the consumption estimates reported in the Burger 2002 study may be
imprecise, in particular for women. The SAB recommends that the Technical Support Document
acknowledge that, while several estimates offish consumption rates are used in the risk assessment,
other estimates reported by Burger could be used. For example, median fish consumption estimates may
better represent the distribution offish consumption data than mean estimates. It should also be
acknowledged that the Burger survey was conducted in 1998, and that fish consumption rates even in
subsistence populations may have changed.
Another issue raised by the SAB focuses on the seasonality offish consumption. Data on consumption
generated from Southern states (e.g., Burger 2002 data from South Carolina) may reflect year-round
consumption, whereas fishers in Northern states may only fish for nine months a year or less. While
failure to take seasonality of fishing into consideration could result in overestimation offish-derived
methylmercury exposure for some regions, the SAB notes that some communities preserve fish for
consumption outside the fishing season. It is important to be certain that fish consumption rates used in
the risk assessment are in the form of annual averages, e.g., consumption rate expressed in terms of
grams offish consumed per week per year. Also, it is unclear whether the risk assessment uses
annualized fish consumption rates and whether fish consumption is based on concentrations that are "as
caught" or "as prepared." The SAB recommends that this information concerning seasonality be
clarified in the Technical Support Document. There is a general agreement that the Technical Support
Document adequately utilizes existing data to identify consumption rates and target populations that are
representative of the most highly exposed susceptible populations.
15
-------
Regarding alternative approaches, the SAB notes that population-based fish consumption rates could be
applied, although these data tend to show lower fish consumption rates than surveys focusing on
subsistence and sport-caught fish (Knobeloch et al. 2005). This would tend to underestimate risks and
would not be consistent with the Technical Support Document objective to target sensitive, highly
exposed individuals. Therefore, this alternative is not recommended.
In regard to fish consumption generally, the SAB recommends that EPA better explain its rationale for
assuming that subsistence consumers eat fish larger than seven inches in length and asks EPA to provide
references supporting its assumptions and to discuss uncertainties associated with this assumption.
38 Use of Census data to identify high-end fish consuming populations
Question 8: Please comment on the approach used in the risk assessment of assuming that a high-end
fish consuming population could be active at a watershed if the "source population "for that fishing
population is associated with that watershed (e.g. at least 25 individuals of that population are present
in aU.S. Census tract intersecting that watershed). Please identify any additional alternative
approaches for identifying the potential for population exposures in watersheds and the strengths and
limitations associated with these alternative approaches (additional detail on how EPA assessed where
specific high-consuming fisher populations might be active is provided in section 1.3 and Appendix C of
the Mercury Risk TSD).
Response: Overall, the SAB agrees that the criterion of using at least 25 persons per census tract from a
given target population (Laotian, poor Hispanic, American Indian populations, amongst others) is a
reasonable approach. The approach is driven by the necessity of using existing data to identify
watersheds with susceptible proximal populations. While the source population selected is somewhat
arbitrary, the SAB agrees that it is a reasonable approach, and that other approaches may not be as
effective or feasible. Regardless of what number is chosen, the prevalence of subsistence fishing in the
target communities remains unknown. EPA indicated that a sample of 25 individuals or greater was
selected to be reasonably certain that at least one subsistence fisher is potentially active at the watershed.
No major concerns are raised by the SAB concerning this issue. However, the SAB recommends that the
Technical Support Document clarify how many census tracts were eliminated due to the use of this cut
point. The Technical Support Document should also include information on the relative distribution of
the sample size of the susceptible populations in the census tracts that were targeted. That is, an absolute
sample of 25 may represent different proportions of the total target population in a given census tract,
which may reflect differences in subsistence fishing behavior. The Technical Support Document should
also discuss the possibility that more remote waterways are fished by subsistence anglers as well and the
potential of this uncertainty for underestimating exposures
3.9. Use of the Mercury Maps approach
Question 9: Please comment on the draft risk assessment's characterization of the limitations and
uncertainty associated with application of the Mercury Maps approach (including the assumption of
proportionality between changes in mercury deposition over watersheds and associated changes in fish
tissue MeHg levels) in the risk assessment. Please comment on how the output ofCMAQ modeling has
been integrated into the analysis to estimate changes in fish tissue MeHg levels and in the exposures and
risks associated with the EGU-relatedfish tissue MeHg fraction (e.g., matching of spatial and temporal
resolution between CMAQ modeling and HUC12 watersheds). Given the national scale of the analysis,
are there recommended alternatives to the Mercury Maps approach that could have been used to link
16
-------
modeled estimates of mercury deposition to monitored MeHg fish tissue levels for all the watersheds
evaluated? (additional detail on the Mercury Maps approach and its application in the risk assessment
is presented in section 1.3 and Appendix E of the Mercury Risk TSD).
Response: Limitations/uncertainty associated with Mercury Maps (MMaps) approach and
proportionality assumption. The risk assessment's qualitative characterization of the limitations and
uncertainty in the application of Mercury Maps approach is appropriate. The SAB recommends that
EPA also summarize quantitative estimates of the uncertainty published in the existing literature in
Appendix F of the Technical Support Document. CMAQ is considered to be the appropriate tool for
providing the link between EGU emissions and mercury deposition to HUC12 watersheds with
methylmercury fish data. There are quite a few comparisons, for example, between mercury wet
deposition as modeled by CMAQ and as observed by the Mercury Deposition Network (e.g., Lin et al.
2007, Prongprueksa et al. 2008, and Bullock et al. 2009). A similar search of the literature for other
components of this risk assessment would allow at least partial quantification of the variability or
uncertainty in this risk assessment, including any literature relating to the time lag in the response of
waterbodies to changes in mercury deposition (e.g., Munthe et al. 2007).
The Mercury Maps model states that for steady-state conditions, reductions in fish tissue concentrations
are expected to track linearly with reductions in air deposition to a watershed with an intercept of zero
for watersheds receiving mercury input exclusively via atmospheric deposition. This proportionality
assumption is extended for the Technical Support Document study so that methylmercury levels in fish
could be apportioned among mercury sources based on the associated apportionment of mercury
deposition within a given watershed. The model is a reduced form of the IEM-2M watershed model used
in theMercury Study Report to Congress (U.S. EPA, 1997), whereby the equations of these models are
reduced to steady state and consolidated into a single equation relating the ratio of current/future air
deposition rates to current/future fish tissue concentrations.
Given these assumptions, Mercury Maps will work only with watersheds in which air deposition is the
sole significant source of mercury and steady-state conditions are assumed. This indicates that the
extension of the proportionality is valid only when other factors influencing methylation potential and
catch profiles (species and trophic levels) remain relatively constant in a given watershed. Watersheds in
which mercury input sources other than air deposition, such as mineral recovery operations using
mercury, mercury cell chloralkali facilities and geologically high mercury inputs, are present and
contribute loads that are significant relative to the air deposition load to that watershed are set aside from
analysis in this risk assessment.
Since the Mercury Maps approach was developed, several recent publications have supported the
finding of a linear relationship between mercury loading and accumulation in aquatic biota (Orihel et al.
2007; Orihel et al. 2008; Harris et al. 2007). These studies suggested that mercury deposited directly to
aquatic ecosystems can become quickly available to biota and accumulated in fish, and that reductions in
atmospheric mercury deposition should lead to decreases in methylmercury concentrations in biota.
These results substantiate EPA's assumption that proportionality between air deposition changes and
fish tissue methylmercury level changes is sufficiently robust for its application in this risk assessment.
Regarding the limitations and uncertainty associated with the application of Mercury Maps, it is
acknowledged that the Mercury Maps approach (i.e., the assumption of proportionality between input
changes and fish response) represents both a critical element of the analysis and a potentially important
source of uncertainty. The sensitivity analyses conducted in the risk assessment addresses two specific
17
-------
uncertainties related to application of Mercury Maps: (1) concerns over including watersheds that may
be disproportionately impacted by non-air mercury sources, and (2) application of the Mercury Maps to
both flowing and stationary freshwater bodies to verify if the two scenarios would produce different
results. The results of these sensitivity analyses suggest that uncertainty related to the Mercury Maps
approach is unlikely to substantially alter the assessment result that mercury emissions from U.S. EGUs
potentially constitute a public health concern.
Integration ofCMAQ data to HUC12 watersheds for estimating changes infishMeHg, exposures and
risks). The use of 12-km spatial resolution in CMAQ modeling is a significant refinement of the
previous analysis, which was conducted using 36-km resolution. The integration of CMAQ data at this
finer resolution into the analysis for estimating changes in fish tissue methylmercury levels is sound,
provided that the proportionality assumption holds true (discussed in the previous response to this
charge question).
CMAQ modeling at a 12-km spatial resolution is used to estimate total annual mercury deposition
caused by U.S. and non-U.S. anthropogenic and natural sources over each watershed. For the purposes
of the risk analysis, watersheds are classified using HUC12 codes (USGS, 2009), representing a fairly
refined level of spatial resolution with watersheds generally 5 to 10 km on a side, which is consistent
with research on the relationship between changes in mercury deposition and changes in methylmercury
levels in aquatic biota. Although interpolating the deposition data from a coarser model grid (CMAQ) to
a finer watershed grid (HUC12) will somewhat diffuse the peak deposition near large point sources, the
data integration approach is sound.
The CMAQ modeling at 12-km resolution is a considerable (nine-fold) spatial refinement of the
modeling conducted to support the Clean Air Mercury Rule (36-km resolution). Modeling results at
finer resolution can be used to better resolve deposition patterns near point sources. The confidence in
applying the 12-km resolution CMAQ results for estimating fish tissue methylmercury changes and its
associated exposure/risk is heavily dependent on the robustness of the proportionality assumption in the
Mercury Maps approach. The limitation and uncertainty of this assumption has been elaborated on in the
response to the first part of this charge question.
Alternatives to the Mercury Maps approach linking modeled deposition to monitoredMeHgfish tissue
levels. The SAB agrees with the application of Mercury Maps in this assessment. There are other
modeling tools capable of making a national scale assessment, such as the Regional Mercury Cycling
Model (R-MCM). However, the R-MCM is more data intensive and the results produced by the two
model approaches should be equivalent.
The R-MCM, a steady-state version of the time-dependent Dynamic Mercury Cycling Model, has been
publicly available to and used by the EPA (Region 4, Athens, Environmental Research Laboratory) for a
number of years. R-MCM requires more detail on water chemistry, methylation potential, etc., and
yields more information as well. Substantial data support the Mercury Maps and the R-MCM steady-
state results, so that the results of the sensitivity analysis and the outcomes from using the alternative
models would be equivalent between the two modeling approaches. Though running an alternative
model framework may provide additional reassurance that the Mercury Maps "base case" approach is a
valid one, it is unlikely that substantial additional insight would be gained with the alternative model
framework.
18
-------
3.10. Exclusion of watersheds with significant non-air loadings
Question 10: Please comment on the EPA 's approach of excluding watersheds with significant non-air
loadings of mercury as a method to reduce uncertainty associated with application of the Mercury Maps
approach. Are there additional criteria that should be considered in including or excluding watersheds?
Response: The technique used to exclude watersheds that may have substantial non-air inputs is sound.
Although additional criteria could be applied, they are unlikely to substantially change the results.
EPA excludes those watersheds that either contained active gold mines or had other substantial non-U.S.
EGU anthropogenic releases of mercury. Identification of watersheds with gold mines is based on a
2005 USGS data set characterizing mineral and metal operations in the United States. The data represent
commodities monitored by the National Minerals Information Center of the USGS, and the operations
included are those considered active in 2003. The identification of watersheds with substantial non-EGU
anthropogenic emissions is based on a Toxic Release Inventory (TRI) net query for 2008 non-EGU
mercury sources with total annual on-site mercury emissions (all media) of 39.7 pounds or more. This
threshold value corresponds to the 25th percentile annual U.S. EGU mercury emission value as
characterized in the 2005 National Air Toxics Assessment. The EPA team considers the 25th percentile
U.S. EGU emission level to be a reasonable screen for additional substantial non-U.S. EGU releases to a
given watershed.
This appears to be a sound approach. The caveat is that TRI reporting may be biased high or low by the
reporting entities, so it is not possible to judge whether the exclusion is reasonably conservative or not.
There is no particular step EPA can take to rectify this uncertainty, although sensitivity tests could be
run on different reporting thresholds and the number (and area) of excluded watersheds that result. As a
minimum, the uncertainty in the TRI should be acknowledged, and the number of watersheds excluded
in the base case and the uncertainty analysis should be explicitly stated.
Other criteria that EPA could consider for exclusion of particular watersheds are:
• Watersheds that are near urban areas, since those may have significant mercury inputs from
runoff which are not included in the TRI reporting database, and
• Watersheds that are excessively polluted, for example by sanitary sewer discharges or highly
anoxic conditions that might deter overall consumer fishing by many users.
3.11. Concentration-response function used in modeling IQ loss
Question 11: Please comment on the specification of the concentration-response function used in
modeling IQ loss. Please comment on whether EPA, as part of uncertainty characterization, should
consider alternative concentration-response functions in addition to the model used in the risk
assessment. Please comment on the extent to which available data and methods support a quantitative
treatment of the potential masking effect offish nutrients (e.g. omega-3 fatty acids and selenium) on the
adverse neurological effects associated with mercury exposure, including IQ loss. Detail on the
concentration-response function used in modeling IQ loss can be found in section 1.3 of the Mercury
Risk TSD.
Response: As noted in the response to questions 2 and 3, the analyses of IQ should assume a less
important role in the final document than in the present one. Question 11 contains three questions
pertaining to the concentration-response function describing methylmercury's effect on IQ. The
response to the first question is that the rationale for the concentration-response function is appropriate,
19
-------
but with qualifications noted below. The response to the second question is that there is no alternative
concentration response function that should be considered, but the analysis should be tempered,
qualitatively, by factors that could influence the shape of the concentration function. The response to the
third question is that masking by fish nutrients could influence the shape of the concentration response
function, but there is not sufficient information to recommend a quantitative adjustment. These three
responses are expanded upon in order below.
The specification of the concentration-response function. The function used comes from a paper by
Axelrad and Bellinger (2007) that seeks to define a relationship between methylmercury exposure and
IQ. A whitepaper by Bellinger (Bellinger 2005) describes the sequence of steps in relating
methylmercury exposure to maternal hair mercury and then hair mercury to IQ. The Technical Support
Document furthers notes that IQ has shown utility in describing the health effects of other
neurotoxicants. These are appropriate bases for examining a potential impact of reducing methylmercury
on IQ, but the SAB does not consider these compelling reasons for using IQ as a primary driver of the
risk assessment. Instead, IQ should serve as a secondary measure along with other measures discussed in
the responses to questions 2 and 3. The modeling of the impact of IQ should be placed in the appendix
and accompanied by the qualifications noted below.
Alternative Concentration Response functions. The concentration-response function derived by Axelrad
and Bellinger (2007) is acceptable for use in supplementary analyses in the Technical Support
Document. It should be noted, however, that this function is likely to underestimate the effect on IQ of
reducing mercury deposition for the reasons itemized here and in the response to charge question 2.
There is another reason that a model based on a linear relationship between exposure and
neurobehavioral effect may underestimate the true effect of reducing exposure. It is evident from animal
studies conducted under highly controlled conditions that the relationship between daily intake and brain
mercury (the most suitable biomarker of exposure) is not linear, but rather is a power function with a
power coefficient that is greater than 1.0; the power coefficient was 1.3 in a review of animal studies
(Newland et al. 2008). This means that a decrease in intake will produce a greater-than-linear decrease
in brain concentration. Thus, the impact of any reductions produced by reducing mercury emissions
could be underestimated by the linear model used in the document. This observation is not intended to
suggest that a new model be used, only that a qualitative argument should be made that the potential
health impact may be underestimated.
A quantitative treatment of the mitigating impact of nutrients. The factors listed in this section could
mitigate the concentration-effect relationship and should be mentioned in the Technical Support
Document, but there is not enough known about their quantitative impact to support a recommendation
of a re-analysis.
There is evidence from the Seychelles study that nutrients can mask effects of prenatal methylmercury
exposures. Davidson et al. (2008), Strain et al. (2008) and Stokes-Riner et al. (2011), demonstrated that
maternal hair mercury was associated with protein disulfide isomerase only after controlling for the
effects of maternal omega-3 polyunsaturated fatty acid (PUFA) status. Controlling for omega-3 PUFAs
steepened the slope of the concentration effect relationship (Strain et al. 2008). These nutrients are found
in many marine fish species, but less is known about their concentration in freshwater fish and the
concentrations may be lower. This issue is important because the concentration-effect relationship used
in the Technical Support Document analysis derives from the consumption of marine fish but it is
applied in the Technical Support Document to the consumption of freshwater fish. Since the slope might
20
-------
be steeper with freshwater fish, it is possible that the analysis in the Technical Support Document
underestimates the impact of reducing mercury deposition on consumers of freshwater fish.
Not only do omega-3 PUFAs mask methylmercury's neurotoxicity, but they confer benefits of their own
that are of direct interest in considering the health impact offish consumption. The studies by Oken et
al. (2005, 2008) directly compared the benefits offish consumption with the hazards associated with
methylmercury exposure. These provide further evidence that the benefits of consuming marine fish
may mask methylmercury's effects, a conclusion that is directly relevant to freshwater fish.
One SAB member points out that methylmercury is a potent inhibitor of multiple families of selenium-
dependent enzymes that are required by the brain and endocrine system (Carvalho et al. 2008; 2011;
Seppanen et al. 2004; Ralston and Raymond 2010). Therefore, the adverse effects of high
methylmercury exposures on these enzymes could be accentuated among populations with poor
selenium nutritional intakes and diminished among those with rich selenium status. Since the
subsistence fish consumers that form the focus of this study are at notable risk of having poor nutrition,
mercury exposures may be non-linearly related to toxicity risks. Other SAB members note that effects of
selenium on methylmercury toxicity are based primarily on observations in animals, and there is
disagreement in the scientific community regarding the significance of these observations to humans.
The same SAB member also suggests that since selenium abundance is largely observed to be inversely
related to mercury bioaccumulation (Chen et al. 2001; Paulsson and Lindberg 1989; Belzile et al. 2004),
diminishments in fish methylmercury concentrations following reductions in mercury deposition will
not be uniform across watersheds. Selenium's inverse relationships to methylmercury bioaccumulation
and toxicity may interact to exacerbate mercury exposure risks in watersheds with low selenium
availability. This SAB member thinks that special consideration should be given to evaluating potential
health risks from consumption offish with high mercury contents that originate from watersheds in low
selenium regions. Other SAB members note that the Mercury Maps (proportional response) approach is
not affected by spatial differences in fish methylmercury content, and in fact this is one of the strengths
of this approach. Changes in fish methylmercury concentrations may differ among aquatic ecosystems in
absolute terms when mercury loading declines, depending upon whether initial fish concentrations are
high or low. However, the reductions in fish methylmercury concentrations within these watersheds are
nonetheless expected to be proportional to the decreases in loading.
Additional Point. Finally a statement on Page 84, Table F-2 references the Seychelles study instead of
the New Zealand study. This should be corrected. The statement is: "Regarding outliers, when an outlier
data point from the New Zealand study was included in the integrated derivation of the IQ loss slope
factor, the factor was reduced by 25 percent (from -0.18 IQ points per unit ppm hair mercury, to -
0.125)." This uncertainty should be acknowledged more explicitly in the body of the document rather
than being merely mentioned in detail in a table in the Appendix. No additional analyses in the
Technical Support Document are necessary; it could just be mentioned in the section on limitations and
uncertainties that risk assessment estimates would be reduced by 25%.
312 Uncertainty and variability
Question 12: Please comment on the degree to which key sources of uncertainty and variability
associated with the risk assessment have been identified and the degree to which they are sufficiently
characterized.
21
-------
Response: To answer this question, the SAB defines variability and uncertainty according to EPA's
standard usage, which is consistent with the definitions given by Cullen & Frey, 1999. These definitions
are as follows:
"Variability refers to temporal, spatial, or interindividual differences (heterogeneity) in
the value of an input. In general, variability cannot be reduced by additional study or
measurement."
"Uncertainty may be thought of as a measure of the incompleteness of one's knowledge
or information about an unknown quantity whose true value could be established if a
perfect measuring device were available."
The Technical Support Document presents a qualitative overview of variability and uncertainty in
Appendix F. The qualitative nature of the discussion is appropriate since this is a conditional analysis.
However, the SAB recommends an expanded discussion in Appendix F of variability and uncertainty to
make explicit the uncertainties associated with the Agency's key analytical choices, which the SAB
supports. This discussion could be organized according to the figures depicting sample calculations of
high and low EGU impact that were provided at the SAB's public meeting on June 15, 2011 and
reproduced below (see Figures 3 and 4, next page). The SAB recommends that these figures be added to
the report along with an explanation of how the calculations were conducted.
Sample Calc-High US EGU Impact (2016)
vl/Fish tissue fppm) 2005 I Example - lake watershed on OH-WV border
(HUC12ID: 50302011009)
IR: daily MeHg intake rate (ug/kg-day)
FTC: MeHg fish tissue concentration (ugJ
FIR: fish ingestion rate (g/day)
CAF: cooking adjustment factor
RfD: methylmercuryRfD
FracEGU: fraction dep from US EGU
Figure 3: U.S. EPA-provided (June 16,2011) schematic showing sample calculation - high U.S. EGU impact
22
-------
Sample Calc-Low US EGU Impact (2016)
:ish tissue (ppm)2005
Mean: 0.27
75th: 0.42
90th: 0.47
5 samples
Example - river watershed on NY
(HUC12ID: 20200040205)
Cooking adjustment
95th% ingestion rate: Factor:
172g/day 1.5
IR: daily MeHg intake rate (ug/kg-day)
FTC: MeHg fish tissue concentration (ug/g or ppm)
FIR: fish ingestion rate (g/day)
CAF: cooking adjustment factor
RfD: methylmercuryRfD
FracEGU: fraction dep from US EGU
US EGU risk = HQ * FracEGU
= 16*0.014 = 0.2 (i.e., <1)
Figure 4: U.S. EPA-provided (June 16,2011) schematic showing sample calculation - low U.S. EGU impact
In addition to the explicit discussion of variability and uncertainty, the SAB suggests that language be
used throughout the Technical Support Document to clarify the scope of the results vis-a-vis variability
and uncertainty in data and methods. For example, the Technical Support Document should cite the
evaluation of uncertainty in the CMAQ and MMAPs source documents. Notwithstanding the
uncertainties in the approach, the SAB considers the approach presented in the Technical Support
Document sound and reasonable.
Variability. The SAB notes the topics covered in Appendix F regarding variability. The clarity of the
documentation of the impact of individual sources of variability could be improved. Carefully selected
maps and additional figures could be particularly helpful in providing this clarity. The SAB recommends
that the following sources of variability be included in Appendix F to avoid misinterpretation of study
results and outcomes.
• The effect of temporal variability in the following on estimates of mercury deposition
o Appendix F should describe CMAQ boundary conditions that are necessary to
establish in order to run the model for the 2 temporal scenarios
• Variation in geographic patterns of populations of subsistence fishers.
o Appendix F addresses geographic variability in total and U.S. EGU-attributable
mercury deposition and fish tissue concentrations. Appendix F should be expanded to
discuss spatial variability in populations of subsistence fishers, noting the limited
geographic coverage of watersheds with fish tissue concentrations.
• Variability in nature and protocols of state collection offish data (see the response to
Question 5, also mentioned below).
• Variation in fisher populations; for example, variation in body weights (potentially across
race/ethnicities) and fishing and consumption habits.
23
-------
• Variability in the factor used to translate mercury concentration measured at time of
collection (i.e., expressed per unit wet weight) in comparison to mercury concentration at
point of consumption following cooking.
Uncertainty. Appendix F defines sources of uncertainty for several components of the overall approach
and selected parameter characterizations. The level of uncertainty is consistent with a screening level
analysis. The SAB advises EPA to strengthen the discussion of each uncertainty presented by
identifying at least qualitatively the direction of its effect on the overall risk assessment. For example,
the small fish sample sizes results in underestimates of the 75th percentiles, which propagates to
conservative underestimates of risk.
The SAB has discussed some sources of uncertainty in responses to other Charge Questions (e.g.,
Question 9). To summarize, the SAB recommends that Appendix F be expanded to provide a more
complete listing and discussion of key uncertainties associated with the assessment. Additional sources
of uncertainty that should be considered for expanded discussion include:
• Overall emission inventories, especially the non-EGU inventory derived as a modified
version of the National Emissions Inventory (NEI). Appendix F should discuss the
uncertainties in inventory components; whether and how the uncertainty changes between the
2005 to 2016 scenarios, including uncertainties in the TRI database; whether there is bias in
the EGU and non-EGU components of the inventory; and whether the EGU emission
estimates are derived from the best performing facilities or from the complete set of facilities.
• Alternative future scenario forecasts. Appendix F should more clearly describe the variables
that are held constant versus factors that are varied between the two scenarios.
• Uncertainty in location of 2016 emissions reductions. Due to EPA's projection methods,
there is uncertainty about where emissions reductions will occur between 2005 and 2016,
which in turn influences the spatial patterns of deposition from EGUs in the 2016 scenario.
Appendix F should address the uncertainties in the 2016 scenario regarding the specific
geographic locations of reductions in EGU-derived mercury deposition as a fraction of total
mercury deposition.
• Use of CMAQ and performance evaluation of CMAQ. Appendix F should provide more
detailed description of uncertainty in CMAQ, including references to existing evaluations of
the model.
• "Hot spots"
o Appendix F should address whether the Mercury Maps approach, as implemented, is
adequate to characterize the existence and extent of mercury "hot spots."
• Impacts of excluding watersheds from the analysis.
o Appendix F should detail the criteria used for excluding watersheds, characterize the
watersheds excluded by different criteria, and describe the estimated deposition in
these watersheds.
• Representativeness of approximately 2,500 watersheds compared to 88,000 HUC12
nationwide.
o Appendix F should characterize any bias introduced by looking at this subset of
watersheds (e.g., some states are over-represented, such as Indiana and Minnesota,
while others are under-represented such as Pennsylvania).
• Fish populations and fish tissue database (see SAB responses to questions 5, 6 and 13 for
more detail). Appendix F should include discussion of:
o Sample size for characterization of Implications of a data set with a low number of
fish per watershed. Appendix F should identify the distribution offish samples per
24
-------
watershed and the possible implications of this distribution, including the
implications of sample size for characterization of 75th percentile fish tissue
concentration.
o Uncertainty in methylmercury fish tissue concentrations from differences in sampling
and analytical protocols used by States that contribute data and errors introduced by
potential misidentification of locations, etc.
• Adjustment between wet and cooked weight offish. EPA relies upon a single older study to
derive an adjustment factor of 1.5. Alternative and newer peer reviewed studies of cooking
effects on mercury in fish should be acknowledged (e.g., Musaiger et al. 2008; Farias et al.
2010) and used to discuss uncertainty associated with this assumed value.
o Appendix F should note that this is a constant value applied in the calculation and
thus does not bias but could skew the results.
• Uncertainty of the assumption of proportionality and the MMAPs approach (see SAB
response to Question 9).
• Characterization of susceptible human populations (see SAB responses to Questions 7 and 8)
o Characterizing subsistence fishing activity within high EGU deposition sites.
o Implications of choosing subsistence fishers and excluding high-end sport fishers.
o Census information that may exclude groups such as students, immigrants).
• Fish consumption rates (see SAB Response to Question 7).
o Limitations of the single study used to support the Technical Support Document's
fish consumption rate for female subsistence fishers.
o Size of fish consumed.
• Derivation of the concentration-response relationship and RfD based on data from marine
fish and mammal species, not inland freshwaters.
o Appendix F should discuss the uncertainty introduced by not using RfDs derived
based on studies of consumption offish from inland freshwaters. (See SAB response
to Question 11).
• Applicability of the concentration-response relationship and RfD for low socio-economic
status populations. This relationship has not been examined.
o Appendix F should discuss how this relationship may bias the report toward
underestimating risk.
• Effect of the nutritional benefits offish consumption in comparison to risks from mercury.
Appendix F should address how the lack of consideration of this factor that may bias the
analysis toward underestimating risk (see SAB response to Question 11).
3.13. Discussion of analytical results
Question 13: Please comment on the draft Mercury Risk TSD 's discussion of analytical results for each
component of the analysis. For each of the components below, please comment on the extent to which
EPA 's observations are supported by the analytical results presented and whether there is a sufficient
characterization of uncertainty, variability, and data limitations, taking into account the models and
data used.
Mercury deposition from U.S. EGUs
Response: EPA's observations in section 2.3 of the Technical Support Document (p. 35)1 are generally
supported by EPA's observations about mercury deposition as depicted in analytical results provided to
Section 2.3 observations:
25
-------
the SAB by EPA following the SAB meeting in the form of a Memorandum from Zachary Pekar, July 1,
2011, entitled "Clarification and Updating of Mercury Deposition Maps Provided in the Technical
Support Document: National-Scale Mercury Risk Assessment." The SAB supports EPA's plan to
include updated figures from the memorandum in EPA's Technical Support Document as replacements
for Figures 2-1 to 2-4 in the March 2011 draft so that they correctly reflect total annual mercury
deposition per square-meter by watershed. The SAB recommends that the spatial patterns of simulated
deposition shown in Figure 2-1 to 2-4 be better explained and that EPA should characterize data
limitations more effectively.
EPA's observations about mercury deposition as depicted in Technical Support Document Figures 2-1
to 2-4 are supported by analytical results. However the 12-km deposition maps are very different than
previously produced maps on the 36-km scale (for example in Texas and Nevada). The SAB
recommends that EPA explain these differences and that EPA consider including separate maps of wet
and dry deposition and/or aggregating the results into an approximately 36 km grid scale for comparison
to earlier maps and to data plots, such as national deposition maps from the Mercury Deposition
Network.
In general, the uncertainties associated with these results are not well characterized or adequately
quantified. For example, there have been several intercomparison studies among numerical models for
long-range transport of mercury and studies on model uncertainty evaluation that are not discussed or
referenced. The SAB recommends that EPA summarize these references (Bullock et al. 2009;
Pongprueksa et al. 2008; Lin et al. 2007; Ryaboshapko 2007) to help frame the overall uncertainty of the
deposition estimates.
In addition, EPA should discuss more completely the inputs that are kept constant for the 2016 scenario
and the inputs that are varied (and by how much). This information may merit discussion earlier in the
report. In addition, the CMAQ results are very dependent on global boundary conditions that are
supplied by the GEOS-Chem model. Uncertainty in those inputs will be carried through to the results.
This should be noted.
Fish tissue methyl mercury concentrations
Response: The observations listed in section 2.4 of the Technical Support Document (pp. 43-44)2 are
generally supported by the analytical results. The SAB recommends that EPA clarify the text to improve
the description of the analytical results for each bulleted observation as described below.
• Patterns of total and U.S. EGU-related Hg deposition differ considerably.
• US Hg deposition is generally dominated by sources other than U.S. EGUs (with the contribution from U.S. EGUs
decreasing between the 2005 and 2016 scenarios).
• The contribution of U.S. EGU deposition to total deposition does vary across watersheds and can represent a
relatively large fraction in some (more limited) instances.
Section 2.4 observations:
• Focus on U. S. EGU-attributable Hg fish tissue levels is in the eastern half of the U.S.
• U.S. EGUs contribute a larger fraction to total Hg fish tissue levels in the U.S. than they do to total Hg deposition
(in terms of percent), this reflects the fact that Hg fish tissue samples are focused in the east where U.S. EGU
deposition is greater.
• Relative to the combined impact of other sources, U.S. EGUs represent a smaller, but still potentially important
contributor to total fish tissue MeHg levels.
• Despite the relatively small fraction of total fish tissue MeHg associated with U.S. EGUs on average, for a subset of
watersheds, they can make a substantially larger contribution.
26
-------
Although there is sufficient characterization of variability, EPA should characterize uncertainty and data
limitations more fully. Specifically, the small sample sizes of mercury concentrations in fish for the
individual watersheds (-29% of watersheds have n=l) will result in lower estimates of mercury
concentrations in the 75th percentile as shown earlier in Figures 1 and 2 in this SAB report. This data
limitation bias will be propagated to underestimate the hazard in the risk assessment.
The text in the observations should be modified to refer to tissue and mercury "concentrations" rather
than "levels" to be more precise. "Level" is a generic term and can refer to any number of different
metrics. Finally, where the percentages of EGU-contribution to fish methylmercury are mentioned, EPA
should clarify that those values pertain to only fish-sampled watersheds. Given the under-sampling in
watersheds where there are high levels of deposition, the percentages indicated could be higher.
Some figures and tables would also benefit from modification or elimination. Figures 2-7 to 2-10 are
difficult to interpret because the symbols do not reflect the number of observations for that site.
Improved plots should display symbols proportional to sample size and provide color or shading of
symbols to represent observed fish concentrations. In addition, the maps shown in Figures 2-7 to 2-14
need to include the western continental United States. These figures unnecessarily cut off the western
continental United States. While the SAB understands the reason for this omission (there is minimal
expected change in EGU emissions in the western United States), it is important to show the results for
the entire United States in the figures of this national assessment. In the absence of national maps, the
reader (especially someone with interest in the western United States) many be left wondering about
current fish methylmercury concentrations in this region (see Figure 2-6), as well as the model predicted
changes in fish methylmercury for the 2016 scenario.
The legend for Figure 2-8 should make it clear that the 2016 mercury tissue concentrations are computed
by adjusting the 2005 concentrations to account for lower expected deposition as per the Mercury Maps
approach. The third bullet item on page 36 of the Technical Support Document should be corrected to
indicate that Figures 2-7 and 2-8 give concentrations of mercury in fish, not total mercury deposition. In
addition, the figures showing the top 10th percentile (2-11 to 2-14) should be removed since the pattern
of mercury is greatly affected by high sampling effort in South Carolina, Indiana, West Virginia and
Louisiana. The current maps could also result in undue public concern in those states. Finally, the text
describing Table 2-5 needs to be clarified to state that the relationships are not causal.
Patterns of mercury deposition with mercury fish tissue data
Response: Overall, the SAB agrees that the observations in section 2.5 of the Technical Support
Document (pp. 4S-49)3 are supported by the analytical results presented and there is a sufficient
characterization of uncertainty, variability, and data limitations. However, a number of changes are
needed to better clarify these points. The Technical Support Document should clearly describe the
degree to which the non-uniform, state-specific data availability influences this analysis. For example,
South Carolina, Louisiana and Indiana all have abundant data availability compared to most states. EPA
should discuss how this data availability bias affects the analytical results. The SAB recommends that
3 Section 2.5 observations:
• The fish tissue MeHg sampling data (summarized at the watershed-level) provides limited coverage for areas with
elevated U.S. EGU Hg deposition. Therefore, the number of "at risk" watersheds as characterized in this risk
assessment may be substantially higher than estimated.
• Hg fish tissue levels are not correlated with total Hg deposition (the relationship is highly dependent on methylation
potential of individual waterbodies).
• Hg fish tissue samples were generally collected in regions with elevated total Hg deposition.
27
-------
this section be substantively rewritten to improve clarity and to highlight the major relevant points. As
discussed below, EPA should revise the text in footnote 36, which is critical to the understanding of
Figures 2-15 and 2-16, and yet is not clearly enough written for the reader to understand the key
information. Also, the figure legends within each of Figures 2-15 and 2-16 need to be changed because
the "blue areas" are not "water bodies," but rather "watersheds," which include water bodies that
sometimes are more obvious than their watersheds (e.g., the Minnesota portion of Lake Superior, Long
Island Sound and perhaps erroneously, the Canadian portion of Lake Champlain). The SAB
recommends that these two maps be replotted with a third color that clearly identifies the areas of
overlap.
Figure 2-17 is critically important not only to this section, but also to the overall document. The SAB
suggests that this figure could be brought into this document much earlier because it adds value to
understanding the lack of direct relationships between deposition and mercury in fish. In a sense, it
frames the justification for the approach taken in the overall analysis. The SAB recommends that EPA
provide a more complete introduction to Figure 2-17 that would state the important premises of the
analysis applied in this risk assessment - that spatial variability of deposition rates is only one major
driver of spatial variability offish methylmercury and that variability of ecosystem factors that control
methylation potential (especially wetlands, aqueous organic carbon, pH, and sulfate) also play a key
role. A question was also raised as to whether Figure 2-17 has been truncated, and if so, did it need to
be? That is, are there data above 1.0 ppm fish concentration and 40 ug/m2-yr deposition? The SAB
suspects that there are.
Figure 2-18 could similarly be moved to an earlier section of the document because it indicates that the
analysis identified watersheds with higher rates of deposition than the national (-88,000 HUC 12
watersheds) trend and that the watersheds with available fish data are in fact, those with higher EGU-
derived mercury deposition rates.
The red areas of Figures 2-15 and 2-16 are labeled in each map's legend as "Watersheds with relatively
elevated US EGU Hg dep." Footnote 36 explains how the red areas are identified, an explanation that is
densely written, as follows:
Footnote 36. Areas of "elevated U.S. EGU-related Hg deposition" refer to areas that are at or above
the average deposition level seen in watersheds with U.S. EGU-attributable exposures above the
MeHg RfD. Specifically, we used exposure estimates based on the 95th percentile fish consumption
rate (for the female high consumer scenario assessed nation-wide) to identify watersheds with U.S.
EGU-attributable exposures above the MeHg RfD and then queried for the average U.S. EGU-
related Hg deposition across that subset of watersheds. This average deposition rate differed for the
2005 and 2016 Scenarios (i.e., 3.79 and 1.28 ug/m2, respectively). These values were used as the
basis for identifying watersheds with levels of U.S. EGU-related Hg deposition for the 2005 and
2016 Scenarios presented in Figures 2-13 and 2-14.
It is troublesome that footnote 36 implies that the threshold for what constitutes "relatively elevated U.S.
EGU Hg deposition" is different in the two maps. The red area in Figure 2-15 is characterized as an
average deposition rate of 3.79 and for Figure 2-16, 1.29 ug/m2. The next, and last, sentence is
confusing, and implies that 3.79 and 1.29 are used as thresholds for identifying the red areas: "These
values were used as the basis for identifying watersheds..." This characterization may confuse readers, in
that readers probably expect similarly colored geographic areas in adjacent similar maps to be presented
as portraying quantitatively similar environmental information, an expectation that these maps
28
-------
apparently do not meet. The SAB suspects that the deposition rate threshold for inclusion in the map is
probably relatively constant, and communicating the threshold would be a more useful characterization
than describing the average deposition rates, which are different for understandable, but unexplained,
reasons. For any given watershed, the threshold is the EGU-attributable deposition rate that produces
EGU-attributable exposure "above the MeHg RfD." In practical terms for this risk assessment of
subsistence fishers, this threshold is a modeled EGU-attributable increment in fish concentration that is
greater than 0.038 ppm methylmercury, a concentration that does not correspond to a constant mercury
deposition rate because the concentration varies among watersheds in accordance with the
proportionality principle described in the risk assessment. However, the SAB notes that the average
mercury deposition rate that produces this incremental methylmercury concentration will be similar
between the 2005 and 2016 scenarios. If so, the red areas could then be characterized, for example, as
"elevated U.S. EGU-related mercury deposition that refers to areas where deposition from EGU
emissions has the potential, even in the absence of mercury from other sources, to cause exposures
above the methylmercury RfD." The average threshold EGU-attributable mercury deposition rate for
exceeding the threshold could be presented, along with the average deposition in the red area. The
revised document should explain why the average deposition rate is lower in the 2016 scenario red area,
rather than assume that the reader will immediately know why.
However the red area is dealt with, a more complete and understandable explanation needs to be
presented than the current explanation of footnote 36.
Percentile risk estimates
Response. Generally, the percentile risk estimates in 2.6.1 are calculated in a reasonable manner and the
observations on pages 53-54 of the Technical Support Document4 are appropriate. The Technical
Support Document especially provides a useful discussion of the uncertainties of high values in Tables
2-5 and 2-7. The SAB has several suggestions to improve the presentation of the material and results for
other parts of section 2.6.1.
The Technical Support Document should include an explanation of why the values in Tables 2-6 and 2-7
decrease when going from the 50th to 75th percentile. This is likely because the ranked risk values are
not the same as the ranked EGU contributions. This difference should be mentioned. Perhaps the tabled
values should be referred to in some way as averaged.
The values in Tables 2-6 and 2-7 are based on averaging the values that are 2.5% below and 2.5%
above. EPA should consider whether it is better to use a 2.5% range or use the 10 nearest values. EPA
should also describe how the range is selected for the 99th percentile.
4 Section 2.6.1 observations:
• For the high-end female consumer assessed at the national-level, total IQ loss and total HQ estimates do not change
in a systematic way between the 2005 and 2016 Scenarios with these levels often being of potential health concern
across a wide variety of consumption rates and watershed percentiles.
• By contrast (again focusing on the high-end female consumer assessed nationally), both U.S. EGU-incremental IQ
loss and the U.S. EGU increment-based HQ display notable reductions between the 2005 and 2016 Scenarios, but
U.S. EGU-attributable risk still exceeds potential levels of concern for a over a quarter of watersheds.
• Estimates of risks generated for the high-end female consumer population (assessed at the national-level) are
generally higher than risks estimated for the other high-end fisher populations, with the exception of white and black
fisher populations assessed in the southeast.
29
-------
In section 2 page 54, the paragraph comparing "risks" for high-end females with other populations is
oversimplified. Depending on the percentiles considered, "risks" for Laotians, Vietnamese and Tribal
fish consumers can also be higher than for high-end females. The highest consumption rates should be
summarized in an appendix.
In section 2 page 55, EPA should provide more information on the gold-mining impacted watersheds in
the Southeast. For example, it seems that gold mining occurred historically in a relatively small region
of South Carolina, and only a few mines have recently been re-activated. Is it really appropriate to
discount or question concerns about EGU affected exposure across the whole Southeast on this basis?
In Tables 2-6 and 2-7, EPA should consider reporting consumption rates and putting the percentiles in
parentheses rather than reporting the percentiles and having the rates in parentheses.
In Table 2-15 and other places, the mean is included. Since the mean is not a percentile, the table header
should be changed or the median used.
Number and frequency of watersheds with populations potentially at risk due to U.S. EGU mercury
emissions
Response: The SAB has no significant concerns regarding the observations in section 2.6.2 of the
Technical Support Document (pp. 57-58).5 The SAB recommends that language be added regarding the
change in the percentage of watersheds that continue to be above the RfD (or above a change in one to
two IQ points, if this aspect of the risk assessment is retained) after EGU emissions are removed.
Furthermore, on the SAB recommends that the first bullet point on page 57 to change the language
"before taking into account deposition..." to something that does not imply temporality (e.g., "when you
factor out other sources of mercury deposition"). The SAB also recommends that if the document
discusses loss of IQ points, the revised document should refer to this change in relation to "populations
living close to watersheds" rather than "watersheds".
With regard to the target population in a broader context, the size of the potentially impacted population
is a key factor to consider in this risk assessment. This issue is outside the scope of the data available for
the risk assessment, even though it is very relevant to the objectives of the Technical Support Document
and its application to public health policy. The document focuses on subsistence fishing populations as a
target population likely to be the most severely impacted by methylmercury consumption in fish. There
is scant evidence documenting the prevalence or extent of subsistence fishing in the United States. Some
SAB members note similarities in consumption rates among sport fishers and subsistence fishing
populations. The inclusion of sport fishers with relatively higher fish consumption rates could expand
the size and extent of the targeted susceptible population. Similarly, only limited information on the
locations or characteristics of watersheds that are excluded from the analysis is provided (p. 63, bullet 4,
Figs 2-15, 2-16). The SAB suggests that more detailed information be included regarding these
watersheds and the uncertainties associated with their exclusion. In addition, the document should
5 Section 2.6.2 observations:
• Less than 1% of the watersheds have an IQ loss of 1 point when deposition from U.S. EGUs is considered before
taking into account deposition and exposures resulting from other sources of Hg.
• Between 2 and 12% of the watersheds haveHQs> 1.5, based on U.S. EGU mercury deposition before factoring in
any other sources of mercury.
• Combining the two categories of watersheds with populations at-risk due to U.S. EGU mercury emissions
summarized in the last two bullets, we get a total estimate ranging from 2 to 28% of watersheds, with this range
reflecting in part the U.S. EGU percent contribution that is considered (e.g., 5, 10, 15 or 20%).
30
-------
address the excluded watersheds within the context of predicted mercury deposition patterns. Some
enumeration of the extent to which the target population would be expanded if these factors had been
incorporated into the analysis would help provide important additional information on the potential
scope and magnitude of the hazards estimated in the assessment. The SAB recognizes that some
additional data may be available on the consumption patterns of recreational anglers, but that EPA did
not have time or resources to integrate this information into the current analysis.
3.14. Responsiveness to the goals of the study
Question 14: Does section 2.8 respond to the goals of the study and does it encapsulate the critical
issues and the significant results of the analysis?
Response: Section 2.8 responds to the goals of the study, but the manner in which it highlights the key
findings could be improved. The section should be revised to explicitly respond to each of the goals of
the study as set out on page 13 of the Technical Support Document:
(a) What is the nature and magnitude of the potential risk to public health posed by current
U.S. EGU mercury emissions?
(b) What is the nature and magnitude of the potential risk posed by U.S. EGU mercury
emissions in 2016 considering potential reductions in EGU mercury emissions
attributable to Clean Air Act requirements? and
(c) How is risk estimated for both the current and future scenario apportioned between the
incremental contribution from U.S. EGU's and other sources of mercury?
In response to these goals, the SAB sees that the major finding of the study is that a reduction in
mercury emissions will translate to reductions in fish tissue methylmercury concentrations, and in turn,
to a reduction in potential risk to subsistence fishers that would result from the consumption of self-
caught fish from inland watersheds. While there are numerous unquantified sources of variability and
uncertainty that are contained in the numerical estimates of potential risk, the variability and uncertainty
do not contradict this basic finding.
3.15. Confidence in the analysis
Question 15: Despite the uncertainties identified, is there sufficient confidence in the analysis for it to
determine whether mercury emissions from U.S. EGUs represent a potential public health hazard for the
group offish consumers likely to experience the highest risk attributable to U.S. EGU?
[Note: This question was not among the original charge questions. It was formulated by the SAB as an
alternative to the second subquestion originally posed by EPA for Charge Question 14, which read as
follows: "/« addition, please comment on the degree to which the level of confidence and precision in
the overall analysis is sufficient to support use of the risk characterization framework described on page
Response; Notwithstanding the uncertainties inherent in this analysis, the Technical Support Document,
after incorporation of the recommendations of the SAB, should provide an objective, reasonable and
credible determination of the potential for a public health hazard from mercury emitted from U.S. EGUs.
31
-------
4. Summary List of Recommendations
This SAB report contains many recommendations for improving the Technical Support Document
presented in the responses to the charge questions. These recommendations can be organized into three
general themes:
1. Improve clarity of the Technical Support Document in terms of the methods used in the risk
assessment and presentation of results. The reader should be able to understand how risk
calculations were performed, the rationale for key decisions regarding the use of models and
sources of input data, and results obtained from the analysis and the conclusions. SAB support
for the risk assessment is contingent upon this improvement in clarity being accomplished.
2. Expand the discussion of sources of variability and uncertainty in the risk assessment. Several
additional sources of uncertainty should be acknowledged and discussed briefly in the Technical
Support document.
3. De-emphasize IQ loss as an endpoint in the risk assessment.
For convenience, specific recommendations have been extracted from the body of the report and are
listed below.
Ouestionl: Overall design
• The Introductory section should make clear, at the earliest possible point, that the analysis is a
determination of watershed impact with exposure addressed as a potential outcome.
Question 2: Critical health endpoints besides 10
• The SAB recommends that EPA reframe the document's discussion of IQ. EPA should
incorporate IQ and other neuropsychological measures as supplemental information and focus on
HQ as the primary critical health endpoint.lt is not suggested that the analyses of IQ be removed
altogether but rather that the analyses be framed in an appendix to the report as a secondary
analysis of impact of reduced exposure on potential health-related outcome. The appendix should
discuss the basis for selecting a HQ at or above 1.5 as the criteria for selecting potentially
impacted watersheds should be explained. The appendix should also include discussion of
potential effects on other measures like developmental delays (Grandjean et al. 1997) or
neuropsychological tests (as discussed by van Wijngaarden et al. 2006), presented in the overall
context of the weight of evidence.
• The SAB recommends that the Technical Support Document acknowledge and discuss
alternative (to HQ) quantitative measures but does not recommend a re-analysis based on these
measures.
Question 4: Spatial scale of watersheds
• In HUCs with multiple lakes, the SAB recommends against using a single fish methylmercury
value to describe the HUC.
• The SAB recommends that the authors provide a summary table describing the characteristics of
the watersheds where fish were collected, including the fraction offish samples collected from
rivers versus lakes, and whether from single or multiple sites.
32
-------
Question 5: Measured fish tissue mercury concentration
• The SAB recommends that the EPA utilize fish methylmercury data collected since 1999 for the
risk assessment.
• The SAB recommends that it be revised to provide a better description of the character of the
data, as well as the selection of analyzable data (e.g., sizes, distribution offish sizes across
watersheds), should be better detailed in the report.
• The SAB recommends that EPA contact some states that receive what the Technical Support
Document terms "relatively elevated" mercury deposition from U.S. EGU emissions and have
limited fish methylmercury measurements to investigate if additional recent (since 1999) fish
methylmercury data are available to improve the coverage for the mercury risk assessment.
Question 6: Use of the 75th percentile fish tissue methylmercury value
• The SAB recommends inclusion of a graph depicting the number of tissue samples available for
analysis by tissue concentration.
• The SAB also recommends that the document discuss this source of uncertainty, including
adding a table with the distribution of number of available fish samples and the fish size from
which they were obtained across watersheds to indicate the extent of the problem. The Technical
Support Document should describe in more detail why including fish tissue concentrations from
one fish sample is likely to result in an underestimate of the number of watersheds at risk.
• The SAB recommends that EPA should also conduct a sensitivity analysis using the median fish
tissue concentration to better represent the distribution offish tissue methylmercury levels where
the sample size is one and provide a bound on the risk assessment.
• The use of other percentiles in the sensitivity analysis is not recommended given the limitations
of the fish tissue data available.
• The SAB recommends that the document describe more clearly the source of the fish
methylmercury data and provide at least a general discussion of how fish sampling programs
differ in ways that can contribute variability and uncertainty to the data set, such as fish capture
methods and criteria for selecting fish to measure methylmercury concentrations.
• The report should include information on the sizes offish that were analyzed. In doing so, the
Technical Support Document may be able to quantify the impact, if any, of the size offish
sampled in watersheds with few fish tissue samples available on estimated mercury
concentrations.
• The SAB also recommends that the Technical Support Document clarify that the 75th percentile
represents available fish tissue data that may or may not represent the fish in the watershed or the
fish consumed.
33
-------
Question 7: Consumption rates and location for high-end consumers
• The SAB recommends that a few caveats should be acknowledged more fully in the document.
The main consumption estimates came from a relatively small survey of individuals attending a
fishing convention in South Carolina, so the consumption estimates reported in the Burger 2002
study may be imprecise, in particular for women. The SAB recommends that the Technical
Support Document acknowledge that, while several estimates offish consumption rates were
used in the risk assessment, other estimates reported by Burger could have been used. For
example, median fish consumption estimates may better represent the distribution offish
consumption data than mean estimates. It should also be acknowledged that the Burger survey
was conducted in 1998, and that fish consumption rates even in subsistence populations may
have changed.
• The SAB recommends that this information concerning seasonality be clarified in the Technical
Support Document
• The SAB recommends that EPA better explain its rationale for assuming that subsistence
consumers eat fish larger than seven inches in length and asks EPA to provide references
supporting its assumptions and to discuss uncertainties associated with this assumption.
Question 8: Use of census data to identify high-end fish consuming populations
• The SAB recommends that the Technical Support Document clarify how many census tracts
were eliminated due to the use of the 25 individual cut point.
• The Technical Support Document should include information on the relative distribution of the
sample size of the susceptible populations in the census tracts that were targeted.
• The Technical Support Document should discuss the possibility that more remote waterways are
fished by subsistence anglers as well and the potential of this uncertainty for underestimating
exposures.
Question 9: Use of the Mercury Maps approach
• The SAB recommends that the quantitative estimates of the uncertainty associated with use of
the Mercury Maps approach published in the existing literature be summarized in Appendix F of
the Technical Support Document.
Question 10: Exclusion of watersheds with significant non-air loadings
• The uncertainty in the TRI (screen) should be acknowledged, and the number of watersheds
excluded in the base case and the uncertainty analysis should be explicitly stated.
Question 11: Concentration-response function used in modeling IQ loss
• IQ should serve as a secondary measure along with other measures discussed in the responses to
questions 2 and 3. The modeling of the impact of IQ should be placed in the appendix and
accompanied by the qualifications discussed in section 3.11 of this SAB report.
• A statement on Page 84, Table F-2 references the Seychelles study instead of the New Zealand
study. This should be corrected.
34
-------
Question 12: Uncertainty and variability
• The SAB recommends an expanded discussion in Appendix F of variability and uncertainty to
make explicit the uncertainties associated with the Agency's key analytical choices, which the
SAB supports. This discussion could be organized according to the figures depicting sample
calculations of high and low EGU impact that were provided at the SAB's public meeting on
June 15, 2011 and reproduced as Figures 3 and 4. The SAB recommends that these figures be
added to the report along with an explanation of how the calculations were conducted.
• The SAB suggests that language be used throughout the Technical Support Document that
clarifies the scope of the results vis-a-vis variability and uncertainty in data and methods. For
example, the Technical Support Document should cite the evaluation of uncertainty in the
CMAQ and MMAPs source documents.
• The SAB recommends that the following sources of variability to be included in Appendix F to
avoid misinterpretation of study results and outcomes.
o The effect of temporal variability in the following on estimates of mercury deposition.
o Appendix F should describe CMAQ boundary conditions that are necessary to establish
in order to run the model for the 2 temporal scenarios.
o Variation in geographic patterns of populations of subsistence fishers.
o Appendix F addresses geographic variability in total and U.S. EGU-attributable mercury
deposition and fish tissue concentrations. Appendix F should be expanded to discuss
spatial variability in populations of subsistence fishers, noting the limited geographic
coverage of watersheds with fish tissue concentrations.
o Variability in nature and protocols of state collection offish data (see the response to
Question 5, also mentioned below).
o Variation in fisher populations; for example, variation in body weights (potentially across
race/ethnicities) and fishing and consumption habits.
o Variability in the factor used to translate mercury concentration measured at time of
collection (i.e., expressed per unit wet weight) in comparison to mercury concentration at
point of consumption following cooking.
• The SAB advises EPA to strengthen the discussion of each uncertainty presented by identifying
at least qualitatively the direction of its effect on the overall risk assessment. For example, the
small fish sample sizes results in underestimates of the 75th percentiles, which propagates to
conservative underestimates of risk.
• The SAB recommends that Appendix F be expanded to provide a more complete listing and
discussion of key uncertainties associated with the assessment. Additional sources of uncertainty
that should be considered for expanded discussion include:
o Overall emission inventories, especially the non-EGU inventory derived as a modified
version of the National Emissions Inventory (NEI). Appendix F should discuss the
uncertainties in inventory components; whether and how the uncertainty changes between
the 2005 to 2016 scenarios, including uncertainties in the TRI database; whether there is
bias in the EGU and non-EGU components of the inventory; and whether the EGU
emission estimates were derived from the best performing facilities or from the complete
set of facilities.
o Alternative future scenario forecasts. Appendix F should more clearly describe the
variables that were held constant versus factors that were varied between the two
scenarios.
o Regarding uncertainty in location of 2016 emissions reductions. Due to EPA's projection
methods, there is uncertainty about where emissions reductions will occur between 2005
and 2016, which in turn influences the spatial patterns of deposition from EGUs in the
35
-------
2016 scenario. Appendix F should address the uncertainties in the 2016 scenario
regarding the specific geographic locations of reductions in EGU-derived mercury
deposition as a fraction of total mercury deposition.
o Use of CMAQ and performance evaluation of CMAQ. Appendix F should provide more
detailed description of uncertainty in CMAQ, including references to existing evaluations
of the model.
o Appendix F should address whether the Mercury Maps approach, as implemented, is
adequate to characterize the existence and extent of mercury "hot spots."
o Appendix F should detail the criteria used for excluding watersheds, characterize the
watersheds excluded by different criteria, and describe the estimated deposition in these
watersheds.
o Regarding representativeness of approximately 2,500 watersheds compared to 88,000
HUC12 nationwide, Appendix F should characterize any bias introduced by looking at
this subset of watersheds (e.g., some states are over-represented, such as Indiana and
Minnesota, while others are under-represented such as Pennsylvania).
o Fish populations and fish tissue database (see SAB responses to questions 5, 6 and 13 for
more detail). Appendix F should include discussion of:
• Sample size for characterization of Implications of a data set with a low number
offish per watershed. Appendix F should identify the distribution offish samples
per watershed and the possible implications of this distribution, including the
implications of sample size for characterization of 75th percentile fish tissue
concentration.
• Uncertainty in methylmercury fish tissue concentrations from differences in
sampling and analytical protocols used by States that contribute data and errors
introduced by potential misidentification of locations, etc.
o Regarding adjustment between wet and cooked weight offish: EPA relied upon a single
older study to derive an adjustment factor of 1.5. Alternative and newer peer reviewed
studies of cooking effects on mercury in fish should be acknowledged (e.g., Musaiger et
al. 2008; Farias et al. 2010) and used to discuss uncertainty associated with this assumed
value.
• Appendix F should note that this is a constant value applied in the calculation and
thus does not bias but could skew the results.
o Regarding uncertainty of the assumption of proportionality and the MMAPs approach
(see SAB response to Question 9 for specifics to be discussed in Appendix F).
o Characterization of susceptible human populations (see SAB responses to Questions 7
and 8)
• Characterizing subsistence fishing activity within high EGU deposition sites.
• Implications of choosing subsistence fishers and excluding high-end sport fishers.
• Census information that may exclude groups such as students, immigrants).
o Fish consumption rates (see SAB Response to Question 7).
• Limitations of the single study used to support the Technical Support Document's
fish consumption rate for female subsistence fishers.
• Size of fish consumed.
o Derivation of the concentration-response relationship and RfD based on data from marine
fish and mammal species, not inland freshwaters.
• Appendix F should discuss the uncertainty introduced by not using RfDs derived
based on studies of consumption offish from inland freshwaters. (See SAB
response to Question 11).
36
-------
o Applicability of the concentration-response relationship and RfD for low socio-economic
status populations. This relationship has not been examined.
• Appendix F should discuss how this relationship may bias the report toward
underestimating risk.
o Effect of the nutritional benefits offish consumption in comparison to risks from
mercury. Appendix F should address how the lack of consideration of this factor that may
bias the analysis toward underestimating risk (see SAB response to Question 11).
Question 13: Discussion of analytical results
Mercury deposition from U.S. EGUs
• The SAB recommends that the spatial patterns of simulated deposition shown in Figure 2-1 to 2-
4 be better explained and that EPA should characterize data limitations more effectively.
• The 12-km deposition maps are very different than previously produced maps on the 36-km scale
(for example in Texas and Nevada). The SAB recommends that EPA explain these differences
and that EPA consider including separate maps of wet and dry deposition and/or aggregating the
results into an approximately 36 km grid scale for comparison to earlier maps and to data plots,
such as national deposition maps from the Mercury Deposition Network.
• There have been several intercomparison studies among numerical models for long-range
transport of mercury and studies on model uncertainty evaluation that are not discussed or
referenced. The SAB recommends that EPA summarize these references (Bullock, 2009;
Pongprueksa et al., 2008; Lin et al, 2007; and Ryaboshapko, 2007) to help frame the overall
uncertainty of the deposition estimates.
Fish tissue methyl mercury concentrations
• EPA should characterize uncertainty and data limitations more fully. Specifically, the small
sample sizes of mercury concentrations in fish for the individual watersheds (-29% of
watersheds have n=l) will result in lower estimates of mercury concentrations in the 75th
percentile as shown earlier in Figures 1 and 2 in this document.
• The text in the observations should be modified to refer to tissue and mercury "concentrations"
rather than "levels" to be more precise.
• Where the percentages of EGU-contribution to fish methylmercury are mentioned, EPA should
clarify that those values pertain to only fish-sampled watersheds. Given the under-sampling in
watersheds where there are high levels of deposition, the percentages indicated could be higher.
• EPA should modify or eliminate some figures and tables.
o For figures 2-7 to 2-10, improved plots should display symbols proportional to sample
size and provide color or shading of symbols to represent observed fish concentrations.
o The maps shown in Figures 2-7 to 2-14 need to include the western continental United
States.
o The legend for Figure 2-8 should make it clear that the 2016 mercury tissue
concentrations were computed by adjusting the 2005 concentrations to account for lower
expected deposition as per the Mercury Maps approach.
o The third bullet item on page 36 of the Technical Support Document should be corrected
to indicate that Figures 2-7 and 2-8 give concentrations of mercury in fish, not total
mercury deposition.
o Figures showing the top 10th percentile (2-11 to 2-14) should be removed since the
pattern of mercury is greatly affected by high sampling effort in South Carolina, Indiana,
West Virginia, and Louisiana.
o The text describing Table 2-5 needs to be clarified to state that the relationships are not
causal.
37
-------
• Patterns of mercury deposition with mercury fish tissue data. The Technical Support Document
should clearly describe the degree to which the non-uniform, state-specific data availability
influences this analysis
• The SAB recommends that this section be substantively rewritten to improve clarity and to
highlight the major relevant points.
• EPA should revise the text in footnote 36, which is critical to the understanding of Figures 2-15
and 2-16.
• Also, the figure legends within each of Figures 2-15 and 2-16 need to be changed because the
"blue areas" are not "water bodies," but rather "watersheds," which include water bodies that
sometimes are more obvious than their watersheds (e.g., the Minnesota portion of Lake Superior,
Long Island Sound, and perhaps erroneously, the Canadian portion of Lake Champlain). The
SAB recommends that these two maps be replotted with a third color that clearly identifies the
areas of overlap.
• The SAB recommends that EPA provide a more complete introduction to Figure 2-17 that would
state the important premises of the analysis applied in this risk assessment - that spatial
variability of deposition rates is only one major driver of spatial variability offish
methylmercury and that variability of ecosystem factors that control methylation potential
(especially wetlands, aqueous organic carbon, pH, and sulfate) also play a key role.
• The revised document should explain why the average deposition rate is lower in the 2016
scenario red area.
Percentile risk estimates
• The Technical Support Document should include an explanation of why the values in Tables 2-6
and 2-7 decrease when going from the 50th to 75th percentile. This is likely because the ranked
risk values are not the same as the ranked EGU contributions. This difference should be
mentioned. Perhaps the tabled values should be referred to in some way as averaged.
• The values in Tables 2-6 and 2-7 are based on averaging the values that are 2.5% below and
2.5% above. EPA should consider whether it is better to use a 2.5% range or use the 10 nearest
values. EPA should also describe how the range is selected for the 99th percentile.
• Section 2 page 54: the paragraph comparing "risks" for high-end females with other populations
is oversimplified. Depending on the percentiles considered, "risks" for Laotians, Vietnamese and
Tribal fish consumers can also be higher than for high-end females. The highest consumption
rates should be summarized in an appendix.
• Section 2 page 55: EPA should provide more information on the gold-mining impacted
watersheds in the Southeast. For example, it seems that gold mining occurred historically in a
relatively small region of South Carolina, and only a few mines have recently been re-activated.
Is it really appropriate to discount or question concerns about EGU affected exposure across the
whole Southeast on this basis?
• In Tables 2-6 and 2-7, EPA should consider reporting consumption rates and putting the
percentiles in parentheses rather than reporting the percentiles and having the rates in
parentheses.
• In Table 2-15 and other places, the mean is included. Since the mean is not a percentile, the table
header should be changed or the median used.
Number and frequency of watersheds with populations potentially at risk due to U.S. EGU mercury
emissions
• The SAB recommends that language be added regarding the change in the percentage of
watersheds that continue to be above the RfD (or above a change in one to two IQ points, if this
aspect of the risk assessment is retained) after EGU emissions are removed.
38
-------
• The SAB recommends that the first bullet point on page 57 to change the language "before
taking into account deposition..." to something that does not imply temporality (e.g., "when you
factor out other sources of mercury deposition").
• The SAB also recommends that if the document discusses loss of IQ points, that it should refer to
this change in relation to "populations living close to watersheds" rather than "watersheds."
Question 14: Responsiveness to the goals of the study
• EPA should revise section 2.8 to explicitly respond to each of the goals of the study as set out on
page 13 of the Technical Support Document.
39
-------
References
Axelrad, D.A., D.C. Bellinger et al. 2007. Dose-response relationship of prenatal mercury exposure and
IQ: an integrative analysis of epidemiologic data. Environmental Health Perspectives 115(4):
609-615.
Bellinger, D.C. 2005. Neurobehavioral Assessments Conducted in the New Zealand, Faroe Islands, and
Seychelles Islands Studies ofMethylmercury Neurotoxicity in Children. Report to the U.S.
Environmental Protection Agency. EPA-HQ-OAR-2002-0056-6045. Available:
http://www.regulations.gov [accessed 20 January 2006].
Belzile, N., Y. Chen, J. Tong, J.M. Gunn, Y. Alarie et al. 2004. The antagonistic role of selenium in
mercury bioassimilation by living organisms, in: 7th International Conference on Mercury as a
Global Pollutant, Volume 51, Pezdic, J., ed., Ljubljana, RMZ-Materials and Geoenvironment,
pp. 803-806.
Bullock, O. R., Jr. et al. 2009. An analysis of simulated wet deposition of mercury from the North
American Mercury Model Intercomparison Study. Journal of Geophysical Researchl 14,
D08301, doi:10.1029/2008JD011224
Burger, J. 2002. Daily consumption of wild fish and game: Exposures of high end recreationalists.
International Journal of Environmental Health Research 12:4, p. 343-354.
Carvalho C.M.L., E.H. Chew, S.I. Hashemy, J. Lu and A. Holmgren. 2008. Inhibition of the human
thioredoxin system: A molecular mechanism of mercury toxicity. Journal of Biological
Chemistry 283 ;18:11913-11923.
Carvalho, C.M.L., J. Lu, X. Zhang, E.S.J. Arner and A. Holmgren. 2011. Effects of selenite and
chelating agents on mammalian thioredoxin reductase inhibited by mercury: Implications for
treatment of mercury poisoning. FASEB Journal 25 (1), pp. 370-381
Chen, Y.W.. N.Belzile and J.M. Gunn. 2001. Antagonistic effect of selenium on mercury assimilation
by fish populations near Sudbury metal smelters. Limnology and Oceanography 46 (7) 1814-
1818.
Chen, C.Y., R.S. Stemberger, N.C. Kamman, B.M. Mayes and C.L. Folt. 2005. Patterns of Hg
bioaccumulation and transfer in aquatic food webs across multi-lake studies in the Northeast US.
Ecotoxicology 14:135-147.
Cullen, A.C. and H.C. Frey. 1999. Probabilistic Techniques in Exposure Assessment. Plenum Press.
Davidson P.W., J.J Strain., G.J.Myers, S.W. Thurston, M.P. Bonham et al.. 2008. Neurodevelopmental
effects of maternal nutritional status and exposure to methylmercury from eating fish during
pregnancy. NeuroToxicol 29: 767-775.
Driscoll, C. T., Y.-J. Han, C. Y. Chen, D. C. Evers, K. F. Lambert, T. M. et al. 2007. Mercury
contamination in forest and freshwater ecosystems in the Northeastern United States. BioScience
57:17-28.
40
-------
Farias, L.A., D.I. Favaro, J.O. Santos, M.B. Vasconcellos, A. Pessoa et al. 2010. Cooking process
evaluation on mercury content in fish. Ada Amazonica 40(4): 741-8.
Grandjean, P., P. Weihe et al. 1997. Cognitive deficit in 7-year-old children with prenatal exposure to
methylmercury. Neurotoxicology & Teratology 19(6): 417-428.
Harris, R. et al. 2007. Whole-ecosystem study shows rapid fish-mercury response to changes in mercury
deposition. PNAS 104 (42) 16586-16591.
Knobeloch L., H.A.Anderson, P. Imm P, D. Peters D and A. Smith. 2005. Fish consumption, advisory
awareness, and hair mercury levels among women of childbearing age. Environmental Research
97(2):219-226.
Lin C.-J., P. Pongprueksa, O.R.Bullock, S.E.Lindberg S.E., S.O.Pehkonen et al. 2007. Scientific
Uncertainties in Atmospheric Mercury Models II: Sensitivity Tests over the Continental United
States. Atmospheric Environment 41, 6544-6560,.
Munthe, I, R.A. Bodaly, B.A. Branfireun, C.T. Driscoll, C.C. Gilmour et al. 2007. Recovery of
Mercury-Contaminated Fisheries. Ambio 36 (1), 33-44.
Musaiger A.O. and R. D'Souza. 2008. The Effects of Different Methods of Cooking on Proximate,
Mineral and Heavy Metal Composition of Fish and Shrimps Consumed in the Arabian Gulf.
Archivos latinoamericanos de nutricion 1(58), 103-109.
Newland, M.C., E.M. Paletzet al. (2008). Methylmercury and nutrition: adult effects of fetal exposure in
experimental models. NeuroToxicology 29(5): 783-801.
Oken, E., J.S. Radesky et al. (2008). Maternal Fish Intake during Pregnancy, Blood Mercury Levels, and
Child Cognition at Age 3 Years in a US Cohort. American Journal of Epidemiology 167(10):
1171-1181.
Oken, E., R.O.Wright, K.P.Kleinman, D. Bellinger, CJ. Amarasiriwardena et al. (2005). Maternal fish
consumption, hair mercury, and infant cognition in a U.S. cohort. Environmental Health
Perspectives 113 (10), 1376-1380.
Orihel, D.M., MJ. Paterson, P.J. Blanchfield, R.A. Bodaly, and H. Hintelmann. 2007. Experimental
evidence of a linear relationship between inorganic mercury loading and methylmercury
accumulation by aquatic biota. Environment, Science & Technology 41, 4952-4958.
http://pubs.acs.org/doi/abs/10.1021/es063061r (accessed 2011).
Orihel D.M., MJ. Paterson, P.J. Blanchfield, R.A. Bodaly, C.C. Gilmour et al.. 2008. Temporal changes
in the distribution, methylation, and bioaccumulation of newly deposited mercury in an aquatic
ecosystem. Environ Pollution 154(l):77-88.
Paulsson, K. and K. Lindbergh. 1989. The selenium method for treatment of lakes for elevated levels of
mercury in fish. Science of the Total Environment 87-88:495-507.
41
-------
Pekar, Zakary. 2011. Memorandum from Zachary Pekar, July 1, 2011, Clarification and Updating of
Mercury Deposition Maps Provided in the Technical Support Document: National-Scale
Mercury Risk Assessment.
http://vosemite.epa. gov/sab/sabproduct.nsf/7EEF811B90F741DF852578C00059FC64/$File/ME
MO-clarification+and+updating+of+mercury+deposition+maps-Julylst.pdf (accessed 9/9/11).
Pongprueksa P., C.-J.Lin, S.E. Lindberg, C. Jang, T.Braverman et al. 2008. Scientific Uncertainties in
Atmospheric Mercury Models III: Boundary and Initial Conditions, Model Grid Resolution, and
Hg(II) Reduction Mechanism. Atmospheric Environment 42, 1828-18451.
Ralston, N.V.C. and L.J.Raymond. 2010. Dietary selenium's protective effects against methylmercury
toxicity. Toxicology 278:112-123.
Ryaboshapko, A.O., O. R. Bullock Jr., J. Christensen, M. Cohen, A. Dastoor et al. 2007.
Intercomparison study of atmospheric mercury models: 1. Comparison of models with short-term
measurements and Intercomparison study of atmospheric mercury models: 2. Modelling results
vs. long-term observations and comparison of country deposition budgets. Science of the Total
Environment 376 228-240.
Seppanen, K., P. Soininen, J.T., Salonen, S. Lotjonen and R. Laatikainen. 2004. Does mercury promote
lipid peroxidation? An in vitro study concerning mercury, copper, and iron in peroxidation of
low-density lipoprotein. Biological Trace Element Research 101, 117-32.
Stokes-Riner, A., S. W. Thurston et al. 2011. A longitudinal analysis of prenatal exposure to
methylmercury and fatty acids in the Seychelles. Neurotoxicology and Teratology 33(2), 325-
328.
Strain J., P.W. Davidson, M.P. Bonham, E.M. Duffy, A. Stokes-Riner et al. 2008. Associations of
maternal long chain polyunsaturated fatty acids, methyl mercury, and infant development in the
Seychelles child development and nutrition study. Journal of Neurotoxicology. 5, 776-82.
U.S. EPA. 1997. Mercury Study Report to Congress. EPA-452/R-97-003
U.S. Environmental Protection Agency Clean Air Scientific Advisory Committee. 2007. Clean Air
Scientific Advisory Committee's (CASAC) Review of the 1st Draft Lead Staff Paper and Draft
Lead Exposure and Risk Assessments (EPA-CASAC-07-003). p. 6.
U.S. Geological Survey and U.S. Department of Agriculture, Natural Resources Conservation Service.
2009. Federal guidelines, requirements, and procedures for the national Watershed Boundary
Dataset: U.S. Geological Survey Techniques and Methods. 11-A3, 55 p.
van Wijngaarden, E., C. Beck et al. 2006. Benchmark concentrations for methyl mercury obtained from
the 9-year follow-up of the Seychelles Child Development Study. NeuroToxicology 27(5): 702-
709.
Watras, C. J., R.C. Back, S. Halrvosen, R.J.M. Hudson, K.A. Morrison et al.. 1998. Bioaccumulation of
mercury in pelagic freshwater food webs. Science of the Total Environment 219:183-208.
42
-------
Wiener, J.G., B.C. Knights, M.B.Sandheinrich, J.DJeremiason, M.E. Brigham et al. 2006. Mercury in
soils, lakes, and fish in Voyageurs National Park (Minnesota): importance of atmospheric
deposition and ecosystem factors. Environmental Science & Technology 40, 6261-6268.
43
-------
Appendix A: Agency Charge Questions
Background and Charge for the SAB Review of EPA's Technical Support Document: National-
Scale Mercury Risk Assessment Supporting the Appropriate and Necessary Finding for Coal- and
Oil-Fired Electric Generating Units (March 2011)
May 23, 2011
Background
On March 16, 2011, EPA proposed National Emission Standards for Hazardous Air Pollutants
(NESHAP) for coal- and oil-fired Electric Utility Steam Generating Units (EGUs). The proposed
NESHAP would protect air quality and promote public health by reducing emissions from EGUs of the
hazardous air pollutants (HAP) listed in Clean Air Act (CAA) section 112(b), including both mercury
and non-mercury HAP. Specifically, the proposed rule would require EGUs to decrease emissions of
mercury, other metal HAP, organic HAP, and acid gas HAP. Section 112(n)(l) of the CAA requires
EPA to determine whether it is "appropriate and necessary" to regulate HAP emissions from EGUs
under section 112. Before the Agency is authorized to make the appropriate and necessary
determination, section 112(n)(l) requires EPA to perform a study of the hazards to public health
reasonably anticipated to occur as a result of HAP emissions, including mercury, from EGUs after
imposition of the requirements of the CAA. EPA completed the required study in 1998. (Utility Air
Toxics Study, 1998). Based in part on the results of that study , EPA made a finding in December 2000
that it was appropriate and necessary to regulate HAP emissions from coal- and oil-fired EGUs. In the
recently proposed NESHAP, EPA confirmed that finding and concluded that it remains appropriate and
necessary to regulate HAP emissions from coal- and oil- fired EGUs. EPA confirmed the finding in part
by conducting a new analysis of the human health risks posed by consuming freshwater fish containing
mercury that is attributable to U.S. EGU emissions of mercury. EPA is seeking peer review of the data
and methods used in the national scale mercury risk assessment as documented in the Technical Support
Document: National-Scale Mercury Risk Assessment Supporting the Appropriate and Necessary Finding
for Coal and Oil-Fired Electric Generating Units (hereafter referred to as the "Mercury Risk TSD").
In determining whether U.S. EGUs pose a hazard to public health, we developed an approach for
assessing the nature and magnitude of the risk to public health posed by U.S. EGU mercury emissions
(the 2005 scenario). We also estimated the health risks associated with US EGU mercury emissions
estimated to remain "after imposition of the requirements of the Act" (the 2016 scenario). Specifically,
for the 2016 scenario, we looked at certain regulations, including, for example, the proposed Transport
Rule, which have a co-benefit impact on mercury.
Our approach focused on identifying the number of watersheds where the U.S. EGU contribution to
total methylmercury (MeHg) risk is considered to represent a potential public health hazard. To do this,
we focused on estimating risk associated with human exposures at those watersheds in the U.S. where
we have measured data on fish tissue MeHg concentrations (about 4% of the watersheds, or 2,461 out of
-88,000 U.S. watersheds - see section 2.4 and Appendix B of the Mercury Risk TSD). For each of the
2,461 watersheds, we modeled potential risk from high-end (i.e., subsistence-level) self-caught fish
consumption. Specifically, we used the fish tissue MeHg data combined with self-caught fish ingestion
rates to model exposure, and then we translated that into estimates of total MeHg-related risk (see
sections 1.3, 2.1 and Appendices C and D of the Mercury Risk TSD).
A-l
-------
In our analyses, we estimated both total risk associated with emissions from all emissions sources,
including global emissions, and the incremental contribution to the total risk that was attributable to
mercury emissions from U.S. EGUs. We used an assumption of proportionality between mercury
deposition over a watershed and the levels of MeHg in fish (and, by association, the levels of exposure
and risk). This proportionality assumption is based on the U.S. EPA Office of Water's Mercury Maps
assessment (see section 1.3 and Appendix E of the Mercury Risk TSD). Mercury Maps demonstrated
that, under certain conditions, a fractional change in mercury deposition will ultimately translate into a
similar fractional change in MeHg levels in fish. We note that the time delay between changes in
deposition and changes in MeHg levels in fish is not well characterized (there are a range of
assumptions and limitations associated with the Mercury Maps approach which we have considered -
see below). Application of the Mercury Maps approach allowed us to translate any changes in mercury
deposition to changes in MeHg fish tissue levels. It also allowed us to apportion MeHg levels in fish
(and, by association, exposure and risk estimates) based on the proportionality assumption. In other
words, if the estimated U.S. EGU-related emissions compriselO% of total deposition over a watershed,
assuming near steady-state conditions are met, we would assume that eventually 10% of the MeHg in
fish (and, therefore, 10% of the total human exposure and risk) would be attributable to U.S. EGUs.
Mercury deposition modeling was completed for two scenarios: 2005 and 2016. The analysis included
consideration of mercury emitted from (a) US EGUs, (b) other non-EGU sources in the U.S. (including
natural and anthropogenic), and (c) sources outside of the U.S. (both anthropogenic and natural) whose
mercury is deposited in the U.S. following long range atmospheric transport. Estimates of mercury
deposition within the U.S., both of total deposition and of EGU-related deposition, were completed
using the Community Multiscale Air Quality model (CMAQ) version 4.7.1, which generates estimates at
the 12 km grid cell-level of resolution.6'7 CMAQ modeling reflects mercury oxidation pathways for both
the gas and aqueous phases in addition to aqueous phase reduction reactions. Mercury "re-emission" is
not explicitly modeled in this version of CMAQ; however, approximations of these emissions are
included in the CMAQ model and called "recycled" emissions. Speciation of U.S. EGU mercury
emissions is based on a factor approach reflecting coal rank, firing type, boiler/burner type, and post-
combustion emissions controls. Emissions of mercury from sources in Canada and Mexico are based on
the 2006 Canadian inventory and 1999 Mexican inventory, respectively. Estimates of mercury
transported into the U.S. from outside North America (i.e., specification of lateral boundary
concentrations, pollutant inflow into the photochemical modeling domain, and initial species
concentrations) are provided by a three-dimensional global atmospheric chemistry model, the GEOS-
CHEM model (standard version 7-04-11). The GEOS-CHEM predictions were used to provide one-way
dynamic boundary conditions at three-hour intervals and an initial concentration field for the 36 km
CMAQ simulations. The 36 km photochemical model simulation is used to supply initial and hourly
boundary concentrations to the 12 km domains.8 Mercury initial and boundary conditions were based on
a GEOS-CHEM simulation using a 2000 based global anthropogenic emissions inventory that includes
1,278 Mg/yr ofHg(O), 720 Mg/yr of Hg(II), and 192 Mg/yr of particle bound mercury.9 The description
6 Foley, K.M., Roselle, S.J., Appel, K.W., Bhave, P.V., Pleim, J.E., Otte, T.L., Mathur, R., Sarwar, G., Young, J.O., Gilliam,
R.C., Nolle, C.G., Kelly, J.T., Gilliland, A.B., Bash, J.O., 2010. Incremental testing of the Community Multiscale Air Quality
(CMAQ) modeling system version 4.7. Geoscientific Model Development 3, 205-226.
7 Byun, D., Schere, K.L., 2006. Review of the governing equations, computational algorithms, and other components of the
models-3 Community Multiscale Air Quality (CMAQ) modeling system. Applied Mechanics Reviews 59, 51-77.
8 USEPA, 2010. Air Quality Modeling Technical Support Document: Point Source Sector Rules (EPA-454/R-11-003),
Research Triangle Park, North Carolina.
9 Selin, N.E., Jacob, D.J., Park, R.J., Yantosca, R.M., Strode, S., Jaegle, L., Jaffe, D., 2007. Chemical cycling and deposition
of atmospheric mercury: Global constraints from observations. Journal of Geophysical Re search-Atmospheres 112.
A-2
-------
of emissions and modeling presented above pertains to the 2005 scenario evaluated in the risk
assessment. For the 2016 scenario, EPA projected US EGU emissions based on an Integrated Planning
Model (IPM) run.10 Mercury emissions from other U.S. anthropogenic sources are projected to 2016
based on growth factors and known controls (e.g., boilers, cement kilns). The estimates for non-U.S.
global emission sources (i.e., both natural and anthropogenic) were not adjusted for the 2016 scenario.
The risk assessment for mercury focuses on two risk metrics: (a) comparison of estimated exposures to
the MeHg Reference Dose (MeHg RfD) to determine the hazard quotient (HQ) for each watershed
evaluated, and (b) an estimate of the number of IQ points lost to children born to mothers exposed to
MeHg during pregnancy (see 1.2 of the Mercury Risk TSD). The current EPA MeHg RfD reflects the
full range of potential neurodevelopmental impacts including effects on IQ, educational development,
motor skills and attention. For the risk assessment, we did not estimate the incidence of adverse health
effects for health endpoints other than IQ loss, as the literature and available data supporting the
modeling of IQ loss is considered to be the strongest and has received the most review by the scientific
community.
For each of the risk metrics modeled (RfD-based HQ and IQ loss), we identified a benchmark for a
potentially significant public health impact to guide interpretation of the risk estimates. For the RfD-
based HQ, we considered any exposure above the RfD (equal to an HQ of 1) to represent a potential
public health hazard with recognition, as noted above, that the RfD provides coverage for the full range
of neruodevelopmental impacts. In the case of IQ loss, we considered a loss of 1 or more points to
represent a clear public health concern. This benchmark was based on advice received from the Clean
Air Science Advisory Committee (CASAC) in relation to the Pb NAAQS review. It is important to note
that CASAC identified this level of IQ loss in the context of a population-level impact (see 1.2 of the
Mercury Risk TSD for additional detail on the benchmarks used to help interpret risk metrics).
For the risk assessment, we focused on high-end (subsistence) fish consumption by women of child-
bearing age at inland fresh water bodies; the consumption rates used ranged from the 90th to 99th
percentiles and were obtained from peer-reviewed studies offish consumption by specific populations
active within the continental U.S. (see section 1.3 and Appendix C of the Mercury Risk TSD). This
overall approach reflects our assumption that U.S. EGUs will have the greatest public health impact on
the subset of watersheds in the U.S. that (a) have relatively elevated fish tissue MeHg levels (increasing
overall risk levels associated with MeHg exposure through fish consumption at those watersheds), (b)
have relatively larger mercury deposition from U.S. EGUs (translating into a greater fractional risk
associated with U.S. EGUs), and (c) have subsistence-level fishing activity (resulting in higher self-
caught fish intake and higher risk). We have not focused on recreational fishing activity. Recreational
fishing may be important from a population risk standpoint; however, these fishers consume less fish
overall and will not have the levels of individual-risk likely to be experienced by subsistence fishers.
Furthermore, we have not considered U.S. EGU impacts on commercial fish from international or near
coastal locations. Although MeHg levels can be relatively high in fish from these locations, the U.S.
EGU contribution (as a fraction of overall mercury impacts) is both highly uncertain and likely to be
low. The high degree of uncertainty associated with linking U.S. EGU deposition to MeHg levels in fish
that are either self-caught or commercially harvested near the U.S. shore led us to exclude consideration
of risks linked to consumption of these fish. Specifically, given the greater mobility of these fish and the
greater dilution of deposited mercury in the ocean and near coastal waters, application of the Mercury
10 USEPA, 2010. Air Quality Modeling Technical Support Document: Point Source Sector Rules (EPA-454/R-11-003),
Research Triangle Park, North Carolina.
A-3
-------
Maps approach is subject to significantly greater uncertainty relative to its application to inland fresh
water bodies.
The RfD-based risk characterization was done by developing HQs for each watershed. The HQ is
defined as the estimate of MeHg exposure divided by the MeHg RfD. Generally (both for
methylmercury and for all pollutants) a HQ of 1 or less is considered to represent a level of daily
exposure for the human population (including sensitive subgroups) that is likely to be without an
appreciable risk of deleterious effects during a lifetime. We developed a 3-stage risk characterization
framework to estimate the number of watersheds where the U.S. EGU contribution to total MeHg risk is
considered to represent a potential public health hazard based on consideration of the HQ metric:
• Stage 1: estimate the number of watersheds where (a) potential exposure for subsistence level fish
consumers exceeds the RfD (e.g., HQ > 1.0), and (b) U.S. EGUs contribute a specific fraction of
mercury deposition to those watersheds (and by association, a specific fraction of total exposure and
risk). Several fractions of mercury deposition were considered ranging from >5 to >20%.
• Stage 2: estimate the number of watersheds where the deposition from U.S. EGUs would result in
exposures to MeHg that exceed the RfD before considering exposures to MeHg attributable to other
sources. While we may consider the U.S. EGU increment of exposure, particularly in the context of
comparing exposure to the MeHg RfD, it is critical to place the U.S. EGU-incremental exposure in
the context of the larger total exposure at a given watershed. This reflects the fact that the MeHg
RfD is for total exposure and not increments of exposure considered in isolation.
• Stage 3: estimate the total number of watersheds where populations are at risk from exposures
attributable to U.S. EGU mercury emissions by merging the two sets of watersheds identified in
stages 1 and 2.
(see section 1.2 of the Mercury Risk TSD for additional detail on the 3-stage framework)
The second risk characterization was done by modeling potential IQ loss attributable to U.S. EGU
emissions resulting in increased MeHg exposure (see section 1.2 of the Mercury Risk TSD). In
modeling IQ loss, we first converted annual-average ingested dose estimates for MeHg into equivalent
maternal hair mercury levels, since the CR function for IQ loss is based on estimated exposure
characterized as maternal hair mercury levels. This was accomplished using a factor based on a one
compartment toxicokinetic model used in deriving the methylmercury RfD. Then a CR function relating
hair mercury levels to IQ points lost in children born to mothers whose exposure is modeled in this
analysis was used to predict IQ points lost for those children. This CR function is based on application
of a Bayesian hierarchical model which integrates data from the three key epidemiological studies
(Seychelles, New Zealand and Faroe Islands).
As part of the risk assessment, EPA also addressed both variability and uncertainty. Regarding
variability, we assessed the degree to which key sources of variability associated with the scenarios
being modeled were reflected in the design of the risk model (see sections 1.4, 2.7 and Appendix F,
Table F-l of the Mercury Risk TSD). Regarding uncertainty we included a number of sensitivity
analyses intended to consider the potential impact of key sources of uncertainty (with emphasis on
application of the Mercury Maps assumption). We also qualitatively discussed additional sources of
uncertainty and the nature and magnitude of their potential impact on risk estimates that were generated
(see section 2.7 and Appendix F, Table F-2 of the Mercury Risk TSD).
Figure 1 provides a conceptual diagram for the key steps in the risk assessment.
A-4
-------
This peer review is intended to focus on the linkages of the key data inputs, and the critical inputs
related to fish consumption rates, dose-response information, and fish MeHg levels. Two key inputs to
the risk assessment are the MeHg RfD and the estimates of mercury deposition from CMAQ. We
believe the MeHg RfD is the appropriate indicator to use because it reflects the full range of potential
neurodevelopmental impacts, including effects on IQ, educational development, motor skills, and
attention. We are not requesting that this panel review the scientific basis for the MeHg RfD, rather, this
review is focused on the estimation of potential exposures to MeHg for comparison against the existing
RfD. The current RfD has been subject to extensive peer review and is the EPA reference value for
assessing MeHg ingestion risk.11 In addition, the CMAQ model has been extensively peer reviewed and
the mercury fate and transport algorithms are documented in several peer reviewed publications.12'13'14
Thus, we are not seeking peer review of the mercury components of the CMAQ model. However, as
reflected in the charge questions, we are looking for comment on how CMAQ outputs (i.e., mercury
deposition estimates) are integrated into the risk assessment to estimate changes in fish tissue MeHg
levels and in exposures and risks associated with the EGU-related fish tissue MeHg fraction.
11 U.S. Environmental Protection Agency (U.S. EPA). 2002. Integrated Risk Information System File for Methylmercury.
Research and Development, National Center for Environmental Assessment, Washington, DC. This material is available
electronically at: http://www.epa.gov/iris/subst/0073.htm.
12 Bullock, O. R., Jr., et al. (2008), The North American Mercury Model Intercomparison Study (NAMMIS): Study
description and model-to-model comparisons, J. Geophys. Res., 113, D17310, doi:10.1029/2008JD009803.
13 Bullock, O. R., Jr., et al. (2009), An analysis of simulated wet deposition of mercury from the North American Mercury
Model Intercomparison Study, J. Geophys. Res., 114, D08301, doi: 10.1029/2008JDO11224
14 Pongprueksa, P., et al (2008), Scientific uncertainties in atmospheric mercury models III: Boundary and initial conditions,
model grid resolution, and Hg(II) reduction mechanism, Atmospheric Environment 42: 1828-1845
A-5
-------
Exposure modeling
Specifying
spatial scale of
watersheds
Characterizing
measured fish tissue
Hg concentrations at
the watershed-level
F
Estimating total fish
consumption-related
Hg exposure forfishers
active at each
watershed
X /
CMAQ
modeling
results fortotal
and US EGU Hg
deposition at
each watershed
1
Apportioning total
exposure between
total and US EGU-
attributable at each
watershed
Defining near-subsistence and
subsistence fisher scenarios
US Demographic data
characterizing source
populations for high-
end fishers at the US
Census tract-level
Defining near-
subsistence and
subsistence fisher
populations by
watershed
Studies characterizing fish
consumption rates for
high-consumption fisher
populations
Risk modeling
Mercury RfD for
HQ estimation
Estimation of HQ and
IQ loss at each
watershed including total
riskand US EGU-
attributable fraction
IV
loc
leling IQ loss
IQ loss concentration-
response function
Factortranslating Hg
ingestion dose into Hg hair
concentrations (in ppm)
KEY:
decision
data input
analysis step
Figure 1. Flow Diagram of Risk Analysis Including Major Analytical Steps and Associated
Modeling Elements (Note, GEOS-CHEM results are input into CMAQ modeling box)
A-6
-------
Charge Questions
The charge questions presented below are organized by topic and track specific sections within the
Mercury Risk TSD beginning with Purpose and Scope of the Analysis (section 1.1). We have included
brief overviews of the technical focus of each section to help reviewers place each section in context
with regard to the overall risk assessment (Note, we did not include any charge questions addressing
elements of the Executive Summary since all technical content provided in that introductory section is
covered in greater detail in the other sections of the TSD for which we have included charge questions).
Purpose and Scope of the Analysis (section 1.1)
This section presents the policy-related questions that were developed to guide the design of the risk
assessment. It also highlights some important technical factors related to air-sourced mercury, in
particular, mercury released from U.S. EGUs that were considered in designing the risk assessment. And
finally, the section provides an overview of key elements of the scope of the risk assessment.
Question 1. Please comment on the scientific credibility of the overall design of the mercury risk
assessment as an approach to characterize human health exposure and risk associated with U.S.
EGU mercury emissions (with a focus on those more highly exposed).
Overview of Risk Metrics and the Risk Characterization Framework (section 1.2)
This section describes the risk metrics used in the risk assessment (i.e., IQ loss and MeHg RfD-based
HQs, including both total risk and U.S. EGU-attributable risk). The section also presents the 3-stage risk
characterization framework which uses these risk metrics to estimate the number of watersheds where
populations may be at risk due to MeHg exposure with consideration for the U.S. EGU attributable
fraction of that exposure. Questions for this section focus on the IQ calculations. As explained above,
we are not asking for peer review of the current mercury RfD or its suitability as a benchmark for
comparison with mercury exposures.
Question 2. Are there any additional critical health endpoint(s) besides IQ loss which could be
quantitatively estimated with a reasonable degree of confidence to supplement the mercury risk
assessment (see section 1.2 of the Mercury Risk TSD for an overview of the risk metrics used in
the risk assessment) ?
Question 3. Please comment on the benchmark used for identifying a potentially significant
public health impact in the context of interpreting the IQ loss risk metric (i.e., an IQ loss of 1 to
2 points or more representing a potential public health hazard). Is there any scientifically
credible alternate decrement in IQ that should be considered as a benchmark to guide
interpretation of the IQ risk estimates (see section 1.2 of the Mercury Risk TSD for additional
detail on the benchmark used for interpreting the IQ loss estimates).
Overview of Analytical Approach (section 1.3)
This section of the Mercury Risk TSD (together with the referenced appendices) provides a detailed
overview of the technical design and inputs to the risk assessment, with the section being further divided
into subsections (unnumbered) that address each of the design elements. Charge questions presented
below which address the design of the risk assessment are grouped by each of these design elements.
A-7
-------
Specifying the spatial scale of watersheds (presented within section 1.3)
This section describes the spatial unit used as the basis for the risk assessment (the HUC-12 watershed,
representing a fairly refined level of watersheds approximately 5-10 km on a side) and provides the
rationale for the decision to use that specific spatial scale and spatial unit in the analysis.
Question 4: Please comment on the spatial scale used in defining watersheds that formed the
basis for risk estimates generated for the analysis (i.e., use of 12-digit hydrologic unit code
classification). To what extent do HUC12 watersheds capture the appropriate level of spatial
resolution in the relationship between changes in mercury deposition and changes in MeHgfish
tissue levels? (see section 1.3 and Appendix A of the Mercury Risk TSDfor additional detail on
specifying the spatial scale of watersheds used in the analysis).
Characterizing measured fish tissue Hg concentrations (presented within section 1.3)
This section describes the fish tissue MeHg sampling data used in the risk assessment, including the
underlying sources of data used in developing the dataset and factors considered in developing the
dataset (e.g., inclusion of data sampled between 2000 and 2009). This section also provides the rationale
for using the 75th percentile fish tissue MeHg value (within a given watershed) as the basis for exposure
and risk characterization.
Question 5: Please comment on the extent to which the fish tissue data used as the basis for the
risk assessment are appropriate and sufficient given the goals of the analysis. Please comment
on the extent to which focusing on data from the period after 1999 increases confidence that the
fish tissue data used are more likely to reflect more contemporaneous patterns of mercury
deposition and less likely to reflect earlier patterns of mercury deposition. Are there any
additional sources offish tissue MeHg data that would be appropriate for inclusion in the risk
assessment?
Question 6: Given the stated goal of estimating potential risks to highly exposed populations,
please comment on the use of the 75th percentile fish tissue MeHg value (reflecting targeting of
larger but not the largest fish for subsistence consumption) as the basis for estimating risk at
each watershed. Are there scientifically credible alternatives to use of the 75th percentile in
representing potential population exposures at the water shed level?
Defining subsistence fisher scenarios (presented within section 1.3)
This section describes the high-end self-caught freshwater fish consuming populations evaluated for
exposure and risk in the risk assessment. The section includes detailed discussion of the self-caught fish
consumption rates used in modeling exposure for these study populations.
Question 7: Please comment on the extent to which characterization of consumption rates and
the potential location for fishing activity for high-end self-caught fish consuming populations
modeled in the analysis are supported by the available study data cited in the Mercury Risk TSD.
In addition, please comment on the extent to which consumption rates documented in Section 1.3
and in Appendix C of the Mercury Risk TSD provide appropriate representation of high-end fish
A-8
-------
consumption by the subsistence population scenarios used in modeling exposures and risk. Are
there additional data on consumption behavior in subsistence populations active at inland
freshwater water bodies within the continental U.S. ?
Question 8: Please comment on the approach used in the risk assessment of assuming that a
high-end fish consuming population could be active at a watershed if the "source population "
for that fishing population is associated with that water shed (e.g. at least 25 individuals of that
population are present in a U.S .Census tract intersecting that watershed). Please identify any
additional alternative approaches for identifying the potential for population exposures in
watersheds and the strengths and limitations associated with these alternative approaches
(additional detail on how EPA assessed where specific high-consuming fisher populations might
be active is provided in section 1.3 and Appendix C of the Mercury Risk TSD).
Apportioning total MeHg exposure between total and U.S. EGU-attributable exposure (presented
within section 1.3)
This section describes the application of the Mercury Maps based proportionality assumption to link
changes in mercury deposition (over watersheds) to changes in fish tissue MeHg levels. The section also
discusses the use of CMAQ modeling output (i.e., gridded mercury deposition estimates for both total
mercury and U.S. EGU-attributable mercury) as part of this process of linking changes in U.S. EGU
mercury emissions ultimately, to changes in fish tissue MeHg levels in watersheds assessed for risk in
the risk assessment.
Question 9: Please comment on the draft risk assessment's characterization of the limitations
and uncertainty associated with application of the Mercury Maps approach (including the
assumption of proportionality between changes in mercury deposition over watersheds and
associated changes in fish tissue MeHg levels) in the risk assessment. Please comment on how
the output of CMAQ modeling has been integrated into the analysis to estimate changes in fish
tissue MeHg levels and in the exposures and risks associated with the EGU-relatedfish tissue
MeHg fraction (e.g., matching of spatial and temporal resolution between CMAQ modeling and
HUC12 watersheds). Given the national scale of the analysis, are there recommended
alternatives to the Mercury Maps approach that could have been used to link modeled estimates
of mercury deposition to monitored MeHg fish tissue levels for all the watersheds evaluated?
(additional detail on the Mercury Maps approach and its application in the risk assessment is
presented in section 1.3 and Appendix E of the Mercury Risk TSD).
Question 10: Please comment on the EPA 's approach of excluding water sheds with significant
non-air loadings of mercury as a method to reduce uncertainty associated with application of the
Mercury Maps approach. Are there additional criteria that should be considered in including or
excluding watersheds?
Estimating risk including HO and 10 loss (presented within section 1.3)
This section describes how exposure estimates generated for the high-end fish consuming populations
modeled in the analysis are translated into risk estimates for those populations (in the form of both
MeHg RfD-based HQs and IQ losses). This section also includes a detailed discussion of the
concentration-response function used in modeling IQ loss.
A-9
-------
Question 11: Please comment on the specification of the concentration-response function used in
modeling IQ loss. Please comment on whether EPA, as part of uncertainty characterization,
should consider alternative concentration-response functions in addition to the model used in the
risk assessment. Please comment on the extent to which available data and methods support a
quantitative treatment of the potential masking effect offish nutrients (e.g. omega-3 fatty acids
and selenium) on the adverse neurological effects associated with mercury exposure, including
IQ loss, (detail on the concentration-response function used in modeling IQ loss can be found in
section 1.3 of the Mercury Risk TSD).
Discussion of key sources of uncertainty and variability (section 1.4)
This section describes the extent to which the risk assessment design reflects consideration for
potentially important sources of variability associated with the type of exposure being modeled. It also
discusses sources of uncertainty associated with the analysis, including the nature and potential
magnitude of their impact on risk estimates (Note, also that an important part of the analysis - the
sensitivity analyses completed primarily to examine the potential impact of uncertainty related to the
Mercury Maps approach - are discussed in section 2.7 of the Mercury Risk TSD).
Question 12: Please comment on the degree to which key sources of uncertainty and variability
associated with the risk assessment have been identified and the degree to which they are
sufficiently characterized.
Discussion of analytical results (section 2)
This section presents estimates generated as part of the risk assessment, including important
intermediate calculations as well as the risk estimates themselves - subsections include: (a) estimates of
mercury deposition over watersheds (section 2.3), (b) characterization of changes in fish tissue MeHg
levels based on modeling the impact of changes in mercury deposition (section 2.4) and (c) presentation
of MeHg RfD-based HQ estimates and IQ loss risk estimates (section 2.6). Key observations from the
analysis are presented in section 2.8.
Question 13: Please comment on the draft Mercury Risk TSD's discussion of analytical results
for each component of the analysis. For each of the components below, please comment on the
extent to which EPA 's observations are supported by the analytical results presented and
whether there is a sufficient characterization of uncertainty, variability, and data limitations,
taking into account the models and data used.
• Mercury deposition from U.S. EGUs
• Fish tissue methyl mercury concentrations
• Patterns ofHg deposition with HG fish tissue data
• Percentile risk estimates
• Number and frequency of watersheds with populations potentially at risk due to U.S. EGU
mercury emissions
Question 14: Please comment on the degree to which the final summary of key observations in Section
2.8 is supported by the analytical results presented. In addition, please comment on the degree to which
the level of confidence and precision in the overall analysis is sufficient to support use of the risk
characterization framework described on page 18.
A-10
-------
-------
United States Office of Air Quality Planning and Standards EPA-452/R-11-009
Environmental Protection Health and Environmental Impacts Division December 2011
Agency Research Triangle Park, NC
------- |